p i from syntactic categories to semantic...

P I

From Syntactic Categories to Semantic Maps

1

Syntactic Argumentation andRadical Construction Grammar

1.1. Introduction

Radical Construction Grammar is a theory of syntax, that is, a theory character-

izing the grammatical structures that are assumed to be represented in the mind

of a speaker. As such, it is broadly comparable to the successive versions of gen-

erative grammar, such as Government and Binding Theory (Chomsky ) and

Minimalism (Chomsky ), and to Head-driven Phrase Structure Grammar

(Pollard and Sag ), Lexical Functional Grammar (Bresnan ), Role and Ref-

erence Grammar (Foley and Van Valin ; Van Valin ; Van Valin and LaPolla

), Functional Grammar (Dik ), Word Grammar (Hudson , ), and

Kay and Fillmore’s Construction Grammar (Fillmore and Kay ; Kay and

Fillmore ; see §.).

One might reasonably ask, does the field of linguistics need yet another

model of syntactic representation? There are more than enough available, in

the eyes of many linguists. Moreover, it is not obvious what the significant

advantages are of any one syntactic theory over the others. All of the aforemen-

tioned theories have developed, in their maturity, a complex array of symbolic

representations of various grammatical structures. The complexity is justified

by their advocates as being necessary to capture the facts of a wide range of lan-

guages, or at least a wide range of facts in a single language. As a result, learning

a syntactic theory is like learning a human language, and in fact it is. A syntactic

theory offers its followers a technical language for describing the structure

of utterances in human languages in presumably insightful ways. If this is true,

then it is important to get the syntactic theory right. But which theory is the right

one?

Radical Construction Grammar differs radically from all of the aforementioned

syntactic theories, as its name implies. Radical Construction Grammar was de-

veloped in order to account for the diversity of the syntactic facts of a single

language as well as the syntactic diversity of the world’s languages. Radical

Construction Grammar emerged from a reassessment of the methods of syntac-

tic argumentation used by linguists since at least the structuralist period. Again,

the reader may reasonably be skeptical. After all, s/he has probably heard such

claims before for other syntactic theories. It is the purpose of this book to per-

suade the reader that Radical Construction Grammar really is different.

Radical Construction Grammar is radical in that it represents a dramatic break

from prior syntactic theories. In this book, I argue that virtually all aspects of the

formal representation of grammatical structure are language-particular. In other

words, Radical Construction Grammar does away with virtually all of the syn-

tactic apparatus that populate other syntactic theories. Thus, the reader will not

find a formal vocabulary for representing syntactic structure in this book, because

such a vocabulary would be language-specific. In this sense, Radical Construction

Grammar is the syntactic theory to end all syntactic theories. There is no longer

any need to learn an entire complex technical language in order to describe a

complex human language.

Radical Construction Grammar is also radical in that it goes back to the foun-

dations of syntax in order to make this new beginning. The reason why we must

discard all of the formal apparatus is a fundamental flaw in the use of the basic

method of empirical syntactic argumentation used in linguistics. This method,

the , defines syntactic categories in terms of their pos-

sibility of filling certain roles in certain . The dis-

tributional method will be described in more detail in §., and the generalized

notion of a grammatical construction will be described in §.. The distributional

method and the linguistic facts it describes turn out to be incompatible with the

assumption that syntactic structures or constructions are made up of atomic

primitive syntactic elements, that is, the elements that make up the building blocks

of other syntactic theories.

Instead, constructions are the basic units of syntactic representation, and cat-

egories are derived from the construction(s) in which they appear—as the distri-

butional method implies. In the rest of this chapter, I explain just how this could

be true. The remaining chapters in Part I apply the argument of this chapter to

two of the most fundamental sets of categories posited by syntactic theories: parts

of speech such as noun and verb, and syntactic roles (“grammatical relations”)

such as subject and object.

The primitive status of constructions does not preclude the possibility of

forming generalizations across constructions and their constituent ,

or of identifying constructions in the first place. In fact, Radical Construction

Grammar presents both of these types of generalizations as instances of the

general cognitive process of (§§..–..). There is a great deal

of research into categorization in psychology and in linguistics, which has revealed

how rich and complex this cognitive ability is. I will briefly touch on these ques-

tions in §.. and §§..–...

The formation of categories is an important aspect of grammatical theory.

However, the reader will not find representations of categorization processes

leading to maximally general analyses of grammatical phenomena in English or

any other specific language in this book. There are several reasons for not exam-


ining categorization in detail in this volume, besides the practical issue of length.

First, a full treatment of this issue must await further psychological research and

its application to language (see the references in §..). Nevertheless, the search

for maximally general categories and rules for a particular language will not yield

what its practitioners believe. Maximally general categories and rules are highly

likely not to be psychologically real (see §..); hence the search for maximally

general analyses is probably a search for an empirically nonexistent—that is, a

fictional—entity. Moreover, constructing a maximally general analysis inevitably

leads to the ignoring of empirical fact—namely, the manifold differences in dis-

tributional patterns of different constructions and categories. In other words,

maximally general analyses are fictional in a second sense, in that they do not

represent empirical reality.

Analyses of maximal generality are often described as “deep” or “detailed”

analyses of grammatical phenomena in a language. A common criticism of

cross-linguistic studies is that they utilize only a “superficial” analysis of each

language. But a genuinely “deep” and “detailed” analysis of a single language would

represent all of the distributional differences among the constructions of the

language. Universals of language will not be found by constructing maximally

general analyses; even if such analyses are not fictional, they are highly language-

specific. Instead, universals of language are found in the patterned variation of

constructions and the categories they define. The careful analysis of variation in

distribution patterns will reveal generalizations about grammatical structure and

behavior in specific languages that are ultimately identical to the cross-linguistic

patterns revealed by typological research, as will be seen repeatedly in this book.

The primitive status of constructions and the nonexistence of primitive syn-

tactic categories is the central thesis of Radical Construction Grammar. This thesis

makes Radical Construction Grammar a theory of syntax, in

contrast to the previously mentioned theories (see §..). But this is not all that

is radical about Radical Construction Grammar.

All of the aforementioned syntactic theories posit the existence of

between the syntactic elements in a construction, such as between a

subject noun phrase and a verb, or between an attributive adjective and a noun,

although the theories vary significantly in exactly how syntactic relations are

represented (e.g. constituency or dependency; see §.. and Chapter ). Radical

Construction Grammar does not posit any syntactic relations in constructions.

The only internal syntactic structure to constructions is their or

part–whole structure: the syntactic structure of constructions consists only of

their elements (which may also be complex constructions) and the that

they fulfill in the construction.

Again, this thesis emerges from a critical examination of methods of syntactic

argumentation. Two classes of evidence are used to argue for the existence of a

syntactic relation between two elements. The first are -

, such as the idiomatic relationship between strings and pull in Strings

Syntactic Argumentation and Radical Construction Grammar

were pulled to get him the job. The second are , such as the

agreement of the Verb sells with the Subject Sheila in Sheila sells seashells.

In Part II, I argue that neither collocational dependencies nor coded depen-

dencies are evidence for the existence of syntactic relations. Collocational depen-

dencies actually reveal between components of the semantic

structure associated with the construction in question (§., following Nunberg

et al. ). Coded dependencies actually reveal between a

formal element of a construction and the semantic component that it stands for

(Chapter ). If the alleged evidence for syntactic relations actually is evidence for

semantic and symbolic relations, then there is no evidence for syntactic relations.

And indeed there are some good reasons not to posit syntactic relations, as will

be seen in Chapter . So Radical Construction Grammar does without them.

Chapter applies the theory expounded in Chapters – to the analysis of heads,

arguments, and adjuncts.

Finally, I argue that constructions themselves are language-specific. In this case,

the method of argumentation that is challenged is the use of a set of necessary

formal syntactic properties for identifying putatively universal constructions

across languages. In Part III, I argue that the facts of languages demonstrate that

any such set of syntactic properties will isolate only a subset of the relevant

constructions in the world’s languages. Instead, one must describe a

in which there is a continuum of construction types in morphosyntactic

terms. I illustrate this approach for voice in Chapter and complex sentences in

Chapter .

Of course, Radical Construction Grammar does not come out of nowhere, and

a number of its central points have been anticipated in other approaches. Cogni-

tive Grammar (Langacker , a, b) is broadly compatible with Radical

Construction Grammar. Cognitive Grammar also eschews a complex syntactic

metalanguage, and relies heavily on the meronomic relations of elements within

syntactic structures (constructions) and on taxonomic relations between con-

structions (see §..). Radical Construction Grammar in turn also conforms to

Cognitive Grammar’s :

the only structures permitted in the grammar of a language (or among the substantive

specifications of universal grammar) are () phonological, semantic or symbolic structures

that actually occur in linguistic expressions; () schemas for such structures; and () cat-

egorizing relationships involving the elements in () and () (Langacker : –).

Radical Construction Grammar engages in a detailed critique of syntactic

methodology and its consequences, and explores cross-linguistic patterns in

greater detail than has been done so far in Cognitive Grammar writings. State-

ments about the differences between Radical Construction Grammar and ‘other

syntactic theories’ in this book should be understood as excluding Cognitive

Grammar from the latter category. Nevertheless, it is important to emphasize that

the arguments on which Radical Construction Grammar rests do not depend on


specific semantic theories or analyses in Cognitive Grammar or elsewhere in the

cognitive linguistic tradition.

Last but not least, Radical Construction Grammar is a theory of syntactic rep-

resentation that is consistent with the findings of field linguists and typological

theory (Comrie , Croft a). The critique of syntactic argumentation to be

found in this book rests in part on the empirical variation of grammatical phe-

nomena, across languages and within them. This empirical variation is well-

known to typologists and in fact to most linguists. Again, it is important

to emphasize that the critique presented here does not depend on the validity of

specific cross-linguistic generalizations proposed in typology. But the question

remains, once the critique has been made, and “Universal Grammar” has been

deconstructed as I have described above, what is left? Where are the universals of

language, if there are any?

The universals of language, if any, are of the type proposed in typological

research. Radical Construction Grammar allows one to bring what might be called

“thinking like a typologist”(Croft ) into the analysis of the grammar of a single

language. There are certain ways of thinking about language that come naturally

to a typologist but which do not always seem to come naturally to other theor-

etical linguists. The following paragraphs briefly describe typological thinking.

Above all, () in language is basic. Variation is the

normal state of language which we have to deal with. It is dealing mainly with

cross-linguistic variation that is the domain of typology. But typologists have

also come to integrate diachronic variation into their purview as in grammat-

icalization theory (Hopper and Traugott ). Radical Construction Grammar

offers a way to integrate synchronic language-internal variation into typological

thinking.

Everything else about doing typology represents typology’s way of dealing with

the fact of variation. A typologist uses an method of analysis, by con-

structing a sample of the world’s languages and seeking language universals via

cross-linguistic generalizations. Since diversity is basic, the only safe way that one

can discover the range of linguistic diversity is by cross-linguistic research. And it

is only through exploring linguistic diversity that one is able to discover the limits

to variation, that is, the universals of human language.

The typologist’s search for language universals is balanced by the recognition

that in language exists and should be accepted as such. Not every-

thing in language can be, or should be, explainable, whether in terms of formal

or functional general principles, abstract generalizations, etc. If it were, all lan-

guages would be alike, all languages would be internally invariant, and no lan-

guages would change.

A typologist endeavors to make his/her language universals explainable, and

hence anything arbitrary about language is, we hope, language-particular. But that

means that the grammars of particular languages involve some arbitrariness

mixed in with the motivated universal principles (Croft a: –). This is


because the principled motivations compete with each other, and the resolution

of the competition for each language is always partly arbitrary. This way of think-

ing is quite foreign to many nontypological linguists, both formalist and func-

tionalist, who seek an explanation for everything in a grammar. Having an eye on

the slightly different language across the ocean, or down the road, generally

reminds the typologist of the futility of this approach (see the Preface).

A typologist also accepts that all things in grammar must pass. Language is

fundamentally , at both the micro-level—language use—and the macro-

level—the broad sweep of grammatical changes that take generations to work

themselves out (Croft a). Synchronic language states are just snapshots of a

dynamic process emerging originally from language use in conversational inter-

action. This thinking follows from the recognition of arbitrariness. What is arbi-

trary can change (since it isn’t dictated by general principles)—and does. What is

basic are the principles that govern the dynamic or diachronic universals.

In fact, anyone who does typology soon learns that there is no synchronic

typological universal without exceptions. But a typologist sees not only the coun-

terexamples—which, after all, must be possible language types, since they actu-

ally exist—but also the highly skewed distribution. In a diachronic perspective,

where every language type comes into existence and passes on to another type

with different degrees of frequency and stability, and the gradualness of change

means all sorts of “anomalous” intermediate types are found, possibility is much

less important than probability. As a result, there has been a shift in typological

thinking from constraining possible language types to calculating probable

language types.1

This, then, is thinking about language like a typologist: variation, arbitrariness,

change, and the fundamentally cross-linguistic character of universals. Radical

Construction Grammar clears the way to bring typological thinking to syntactic

theory.

Radical Construction Grammar does not merely aim to deconstruct syntax as

we know it. Another major goal of Radical Construction Grammar is to represent

universals of human language in a way that is plausible as a model of the know-

ledge of an individual speaker of a particular language. Recently, typologists have

begun to embrace a model of the representation of language-particular gram-

matical knowledge in the context of universal patterns of variation. This model

is the model, in which distributional patterns of language-

particular categories are mapped onto a , much of whose

structure is hypothesized to be universal. The semantic map/conceptual space

model will be introduced in §., and used throughout Parts I and III of this book

(Part II is devoted to the internal syntactic structure of constructions).


1 Dryer offers powerful arguments for why probabilistic (‘statistical’) language universals are

superior to categorical (‘absolute’) language universals, given that the known distribution of the vast

majority of grammatical properties is skewed, often strongly skewed.

1.2. Methodology and Theory in Syntax

1.2.1. The basic question of syntactic analysis

Linguistics is an empirical science. That is to say, any syntactic theory must be

developed in a conversation with the facts of human languages. Hence, there are

two intertwined basic questions of syntactic analysis, one theoretical and one

methodological.

The theoretical question posed by contemporary syntactic theories is: ‘

⁄ , ?’ That is, what is the right

theory of syntax, more precisely, of syntactic representation—in particular one

that conforms to the empirical discoveries of cross-linguistic research?

The theoretical question is commonly cast as the search for U

G. The term ‘Universal Grammar’ and its abbreviation, UG, are associ-

ated with generative grammar’s claim that certain aspects of syntactic structure

are not only universal but innately specified. However, one can characterize Uni-

versal Grammar more broadly, without necessarily assuming that it is innate:

although innate properties are necessarily universal, universals of human language

are not necessarily an innate genetic endowment. In the broad sense, Universal

Grammar consists of those aspects or properties of grammatical structure which

form the makeup of the grammars of all human languages.

The syntactic theories named in §. offer a precise answer to this theoretical

question, in terms of the formal language for syntactic representation that each

theory provides. Universal Grammar is the formal language for syntactic repre-

sentation. These syntactic theories are generally called theories of

syntax. Formalist theories of syntax are generally contrasted with

theories of syntax. Functionalist theories of syntax are associated with the hypo-

thesis that the principles that govern syntactic structure are not self-contained,

that is, they involve semantic and/or discourse principles. In contrast, formalist

theories argue that there is a set of self-contained principles governing syntactic

structure that make no reference to semantic or discourse principles.

It is not my purpose to discuss the self-containedness of grammar here (see

Croft a). Instead, I wish to focus on a fact about grammar accepted by both

formalists and functionalists (see Croft a: , ): that the form–function

(syntax–semantics) mapping is to at least some degree, and thus form

must be represented independently of function to at least some degree. Formal-

ist theories of syntax incorporate the arbitrariness of language in their models

fairly straightforwardly. Functionalists, on the other hand, have rarely addressed

the question of how the formal structure of utterances is to be represented (two

important exceptions are Langacker and Givón ). This is the question

that this book addresses.

All of the syntactic theories mentioned in §. share a fundamental assumption

about the nature of Universal Grammar. It is assumed that Universal Grammar


consists of a set of atomic grammatical primitives (categories and relations), over

which syntactic structures and constraints on those structures are defined, that

are the building blocks of the particular grammars of all human languages. This

assumption is shared with traditional grammar, which gave us much of the ter-

minology used to describe these atomic grammatical primitives, such as ‘noun’,

‘verb’, ‘subject’, and ‘object’. This assumption is presented as fact in introductory

linguistics textbooks, such as O’Grady et al. (), and in introductory syntax

textbooks, such as Haegeman ():

A fundamental fact about words in all human languages is that they can be grouped

together into a relatively small number of classes, called syntactic categories (O’Grady et

al. : ).

Words belong to different syntactic categories, such as nouns, verbs, etc., and the syntac-

tic category to which a word belongs determines its distribution, that is, in what contexts

it can occur (Haegeman : ).

Syntactic theories have qualified the universality of Universal Grammar in some

cases. A strict interpretation of Universal Grammar would assert that all languages

possess essentially the same set of categories and relations. A

interpretation of Universal Grammar asserts that the presence of some properties

are contingent on the presence of other properties. It is usually assumed that a

biconditional universal relation holds between the parameterized properties, such

that if a language has +a then it has +b, +g, etc. and if a language has -a then it

has -b, -g, etc. Finally, a “” interpretation of Universal Grammar

asserts that the set of categories and relations are available to all speakers, but

speakers of some languages do not avail themselves of all of the categories and

relations available to them (e.g. Zwicky : , fn ). For example, one approach

to the category Aux(iliary) proposed that Aux was a part of Universal Grammar,

but not every language had a category Aux. These qualifications are intended to

accommodate some of the diversity of human languages, but they do not signifi-

cantly alter the basic reductionist model.

The theoretical question is the one emphasized in most textbooks and presen-

tations of syntactic theories. But in empirical science, there is always a method-

ological question that is presupposed by the theoretical question. For syntactic

theory, the methodological question is: ‘ , -

— —

?’

If we propose that adjective or subject are categories in Universal Grammar, we

must have a method for deciding whether a particular language has adjectives or

subjects. Likewise, if we construct a hypothesis about, for instance, an essential

grammatical property of subjects across languages, then we must be able to iden-

tify subjects in every language in order to observe if the subjects have the gram-

matical property in question.

It has been suggested to me that the methodological question is of relatively


minor importance. In particular, reference is made to Chomsky’s argument that

it is unreasonable to ask linguistic theory for a discovery procedure for identify-

ing the right grammar for a particular language (Chomsky : –). However,

the problem which I am referring to is more basic than that. It is what Chomsky

calls the condition of generality (ibid., ), necessary for any adequate theory of

grammar: ‘we must characterize the form of grammars in a general and explicit

way so that we can actually propose grammars of this form for particular lan-

guages’ (ibid., –). That is, for a particular language we can argue for and thus

justify the analysis of that language’s structures as an instance of the structures

found in Universal Grammar.

It is the condition of generality that I believe current syntactic theories fail. That

is, the methods that linguists use to argue for their syntactic theories carry hidden

fallacies which are largely unremarked upon. When these fallacious assumptions

are uncovered, their abandonment leads us to a very different approach to syn-

tactic theory than that advocated by formalist theories and even the functionalist

syntactic theories referred to above.

1.2.2. Distributional analysis: the basic method of syntactic argumentation

There is essentially one single, simple, method that is widely used for syntactic

analysis. This method is called . Distributional analysis

stretches far back in the history of linguistics; but it was first codified and given

its name by American structuralist linguists in the middle of the last century:

Descriptive linguistics, as the term has come to be used [i.e. structural linguistics, cf. Harris

: ], is a particular field of inquiry which deals . . . with the regularities of certain fea-

tures of speech. These regularities are in the distributional relations among the features of

speech in question, i.e. the occurrence of these features relatively to each other within utter-

ances (Harris : ).

In distributional analysis, syntactic categories are defined by the occurrence or

nonoccurrence of their members in different types of utterances. Utterance types,

such as the Information Question (What did you see?) or the Passive (The bride

was greeted by the guests) are defined in structural terms. We will call utterance

types ; constructions will be discussed in greater detail in §..

Also, members of a syntactic category do not occur just anywhere in a construc-

tion; their occurrence in a construction is relative to the they fill in the con-

struction; for example, The bride fills the Subject role in the Passive construction.

Distributional analysis can be illustrated with a couple of simple examples.

Consider the occurrence/nonoccurrence of the English words cold, happy, dance,

and sing in examples –:2


2 In these examples and most other examples in this book, the salient morphosyntactic features dis-

cussed in the text are highlighted by boldface. This will, I hope, allow the reader to follow the argu-

ments in the book, especially when examples from little-known languages are used as evidence

supporting the argument. In some examples (e.g. – below), the original language form conflates

different grammatical categories; in these case, only the relevant part of the gloss is put in boldface.

In Table ., the columns represent the constructions illustrated in the (a) and

(b) sentences of – respectively. The construction is represented by specifying the

role of each element in the construction. If the role represents a category, the cat-

egory label is given in small capitals; if the role can be filled by only a specific

word or morpheme such as be, that word or morpheme is given in italics. The

role that is being examined is represented by __. The internal syntactic structure

of the construction, in particular the order of elements, is not further specified

here (see §.. and Part II for further discussion).

The rows represent the words in – that occur or do not occur in the speci-

fied roles in the constructions. Occurrence is indicated by √ and nonoccurrence

by *. In Table ., the words with the same pattern of occurrence/nonoccurrence

are put in a single row. The result is two classes of words, which we have labeled

with their traditional names, Adjective and Verb. In this book, we will use the con-

vention of capitalized names for language-specific categories, following Comrie

a and Bybee (see §..). The difference in distribution of the words in

– in the two constructions is taken to be evidence of the existence of the syn-

tactic categories Adjective and Verb in English.


Table .. Distribution of Verbs and Adjectives in predicate

constructions

[S be __] [S __-.]

Adjective: cold, happy, etc. ÷ *

Verb: sing, dance, etc. * ÷

() a. Jack is cold.

b. *Jack colds.

() a. Jack is happy.

b. *Jack happies.

() a. *Jack is dance.

b. Jack dances.

() a. *Jack is sing.

b. Jack sings.

In the (a) sentences, the four words occur in the Predicate role after the inflected

copula be. Sentences a and a are acceptable, but a and a are not. In the (b)

sentences, the four words occur in the Predicate role without a Copula but with

the Tense Agreement inflection (in this case, rd Person Singular Present Tense -s).

Here, sentences b and b are not acceptable, but sentences b and b are. Thus,

cold and happy have the same distribution, and contrast with dance and sing,

which share a different distribution pattern. The overall distribution pattern is

given in Table ..

The pattern of occurrence/nonoccurrence of the words in each row across the

constructions in the columns in Table . is called the or the -

of the word(s) in question. Each construction that is used to define a syntac-

tic category in this way is typically called an , , or for

the category in question. Thus, is the use of one or

more constructions to justify the existence of a particular category in the grammar

of a language.

What is the explanation of the distribution of the syntactic categories in Table

.? The quotation from Haegeman () in §.. suggests that it is the category

membership of the words in question. A purely semantic explanation will not

work for parts of speech (see Chapter ), and so a syntactic explanation is given

instead, in terms of the category membership of the words in question. This ex-

planation is considered to be stronger if one can show that the category in ques-

tion occurs across many constructions. For example, it has been argued that the

categories Subject and Object as syntactic roles are found in several different con-

structions (examples adapted from Croft a: , following standard arguments

in the syntactic literature; see also §..):

() Nominative form of the Pronoun:

a. She congratulated him.

b. *Her congratulated he.

() Agreement of the Verb:

a. Shei likesi horses.

b. *She like-Øj horsesj.

() “Null” Noun Phrase in Infinitive Complement “controlled” by Main Clause Subject:

a. Jack told Fredi Øi to buy a car.

b. *Jack told Fredi to give Øi $.

() “Null” Noun Phrase in Imperatives:

a. Ø Learn Hungarian!

b. *Mary teach Ø Hungarian!

() “Null” Noun Phrase in Conjunction Reduction Coordination (see e.g. McCawley :

):

a. Shei fell and Øi broke her hip.

b. *Shei died and they buried Øi.

The arguments for Subject vs. Object in English are summarized in Table .

(p. ).

In standard syntactic argumentation, the more constructions that appear to

include a particular category as a role, the stronger the evidence for that category

in the language is. That is, the more arguments for a syntactic category that can

be offered, the better.

The distributional method is the basic method of empirical grammatical analy-

sis. It is used to identify the basic grammatical units out of which complex syn-

tactic structures or constructions are built. However, the application of the

distributional method reveals problems that can only be resolved by taking con-


structions—complex syntactic structures and their meanings—rather than cat-

egories as the basic units of grammatical representation. Before discussing the

problems with the distributional method, I will describe what constructions are

and how they are represented in construction-based syntactic theories.

1.3. Constructions and Construction Grammar

In this section, I will give a brief introduction to constructions and construction

grammar; this introduction is based on a more detailed explication in Cruse and

Croft (to appear, chapters –). In §.., I will give an outline of some of the

arguments in favor of a construction grammar approach to syntactic theory.

Needless to say, this section can give only a summary of the more important argu-

ments, and the curious reader should examine the more detailed arguments in

Fillmore, Kay and O’Connor , Goldberg , Kay and Fillmore and

Cruse and Croft (to appear). In §.., I describe the internal structure of a con-

struction. In §.., I briefly discuss the organization of constructions in a

grammar.

Construction grammar exists in a number of variants, such as those found in

Lakoff (), Fillmore and Kay (; see also Kay and Fillmore ), Goldberg

() and Langacker (, ). Radical Construction Grammar, as its name

implies, is another variety of construction grammar. The introduction here

emphasizes the commonalities among the different models of construction

grammar, and the particular points where Radical Construction Grammar will be

seen to differ from the other construction grammars.

1.3.1. Arguments for construction grammar

Construction grammar represents a reaction to the of the

organization of a grammar that is found in other syntactic theories. In the com-

ponential model, different types of properties of an utterance—its sound struc-

ture, its syntax and its meaning—are represented in separate components, each


Table .. Traditional constructional tests for subjecthood

in English

I II III IV V

Subject ÷ ÷ ÷ ÷ ÷Object * * * * *

I: Nominative Case marking (Pronouns)

II: Agreement of the Verb

III: Controlled null NP of Infinitive Complement

IV: Null NP of Imperative

V: Null NP in “Conjunction Reduction” Coordination

Many current theories in fact divide grammatical properties into a larger

number of components, e.g. a morphological component, an information struc-

ture component, or a variety of syntactic components. However these modifica-

tions still adhere to the basic concept behind the componential model: gram-

matical properties of different types are placed in separate components, except

for the lexicon.

Construction grammar arose out of a concern to analyze a problematic phe-

nomenon for the componential model, namely idioms (Fillmore et al. ).

Idioms are linguistic expressions that are syntactically and/or semantically idio-

syncratic in various ways, but are larger than words, and hence cannot simply be

assigned to the lexicon without some special mechanism. Some idioms are lexi-

cally idiosyncratic, using lexical items found nowhere else, such as kith and kin

‘family and friends’. Such idioms are by definition syntactically and semantically

irregular, since the unfamiliar word has no independent syntactic or semantic

status. Other idioms use familiar words but their syntax is idiosyncratic, as in all

of a sudden or in point of fact ; these are called idioms. Still

other idioms use familiar words and familiar syntax but are semantically idio-

syncratic, such as tickle the ivories ‘play the piano’.

A theory of grammar should of course capture the differences among these

types of idioms and their relationship to the regular lexicon and regular syntac-

tic rules of a language. The need for a theory that can accommodate idioms is

even more critical for the idioms that Fillmore et al. () focus their attention

on, idioms which are to a greater or lesser degree. That is, some idioms

are not completely lexically specific or , like the idioms in the pre-

ceding paragraph, but instead include whole syntactic categories admitting a wide

range of possible words and phrases to instantiate those categories.


Figure .. The componential model of the organization of a grammar

of which consists of rules operating over primitive elements of the relevant types

(phonemes, syntactic units, semantic units). The only constructs which contain

information cutting across the components are words, which represent conven-

tional associations of phonological form, syntactic category, and meaning. More

recently, attention has been directed to that link complex syntac-

tic structures to their semantic interpretation, and link syntactic structures to their

phonological realization. The componential model is illustrated in Figure ..

Partially schematic idioms also range over all three types described by Fillmore

et al. A schematic idiom which is lexically idiosyncratic is the comparative con-

ditional construction The X-er, the Y-er as in The longer you practice, the better you

will become (the form the is not directly related to the definite article, but is derived

from the Old English instrumental demonstrative form y). An example of an

extragrammatical schematic idiom is the “cousin” construction Nth cousin (M

times removed), as in second cousin three times removed, which describes different

kinds of distant kin relations and has its own unique syntax. Finally, an example

of a schematic idiom that is only semantically idiosyncratic is pull NP’s leg ‘joke

with NP’ as in Don’t pull my leg; the NP category can be filled by any Noun Phrase

denoting a human being.

Schematic idioms pose a serious challenge to the componential model because

schematic idioms either have regularities of their own which ought to be captured

as regularities (the extragrammatical schematic idioms), or follow regular syn-

tactic rules and ought to be somehow represented as doing so (the grammatical

schematic idioms). Moreover, all idioms are semantically idiosyncratic, which

means that they do not follow general rules of semantic interpretation. Instead,

they have their own rules of semantic interpretation.

Fillmore et al. () argue that we should accept the existence of idioms as

. Constructions are objects of syntactic representation that also

contain semantic and even phonological information (such as the individual sub-

stantive lexical items in the partially schematic idioms, or special prosodic pat-

terns or special rules of phonological reduction as in I wanna go too). In other

words, constructions are like lexical items in the componential model: they link

together idiosyncratic or arbitrary phonological, syntactic, and semantic infor-

mation. The difference between lexical items and constructions is that lexical

items are substantive and (that is, minimal syntactic units), while con-

structions can be at least partially schematic and (consisting of more

than one syntactic element).

Beginning with Fillmore et al. () and Lakoff (), there have been a

number of detailed studies of constructions whose grammatical properties cannot

be accounted for by the general syntactic, semantic, and pragmatic rules of

English; other major studies following Fillmore et al. and Lakoff ’s model include

Goldberg () and Michaelis and Lambrecht (). Also, the studies of syn-

tactic structures with special pragmatic functions by Prince () and Birner and

Ward (), and the studies of syntactic structures with special semantic inter-

pretations by Wierzbicka (, , ), strengthen the case for treating those

syntactic structures as constructions. Even formal syntacticians who adhere to the

componential model have recognized the existence of constructions to some

extent; see for example Akmajian () (and compare Lambrecht’s reanaly-

sis of the same phenomenon) and Jackendoff (a, ).

Constructions can be thought of as the same theoretical type of representation

object as lexical items, albeit syntactically complex and at least partially schematic.

þ


Thus, there is a between the lexicon and syntactic constructions. Fill-

more et al. () also make the logical next step: regular syntactic rules and

regular rules of semantic interpretation are themselves constructions. The only

difference between regular syntactic rules and their rules of semantic interpreta-

tion and other constructions is that the former are wholly schematic while the

latter retain some substantive elements.

The constructional tail has come to wag the syntactic dog: everything from

words to the most general syntactic and semantic rules can be represented as con-

structions. The final step is to recognize that the internal structure of words are

also constructions. After all, a construction like [The X-er, the Y-er] or [pull-

NP-’s leg] includes bound morphemes and/or clitic elements in its syntactic rep-

resentation. One can have fully morphological constructions such as [N-s]

(partially schematic) or [V-] (wholly schematic), while an individual word

form like child-ren is a wholly substantive morphological construction. The only

difference between morphological constructions and syntactic ones is that the

former are entirely made up of bound morphemes while the latter are largely

made up of free morphemes.

In other words, construction grammar has generalized the notion of a con-

struction to apply to any grammatical structure, including both its form and its

meaning. The logical consequence of accommodating idioms in syntactic theory

has been to provide a uniform representation of all types of grammatical struc-

tures from words to syntactic and semantic rules. The uniform representation is

referred to as the – (compare Langacker : –,

–), illustrated in Table ..


Table .. The syntax–lexicon continuum

Construction type Traditional name Examples

Complex and (mostly) schematic syntax [S be- V-en by O]

Complex and (mostly) specific idiom [pull- NP-’s leg]

Complex but bound morphology [N-s], [V-]

Atomic and schematic syntactic category [D], [A]

Atomic and specific word/lexicon [this], [green]

Construction grammar’s great attraction as a theory of grammar—not just

syntax—is that it provides a uniform model of grammatical representation and

at the same time captures a broader range of empirical phenomena than compo-

nential models of grammar. For the same reason, construction grammar also pro-

vides the most general and neutral way to describe the distributional method

(§..). The notion of a construction in construction grammar is broad enough

to represent any morphological or syntactic arguments/criteria/tests for identify-

ing any syntactic category. For example, a description of a distributional

argument or test characterized as applying a syntactic or morphological rule to

the item in question (such as the Passive rule) can always be recast as a descrip-

tion in terms of occurrence of that item in the construction that is described by

the output or result of applying the rule (the Passive construction).

The generality and empirical coverage of construction grammar is a major

attraction of construction grammar. However, the arguments in favor of Radical

Construction Grammar do not presuppose construction grammar. The argu-

ments to be presented in this chapter and in Chapters – are methodological

arguments and thus apply to other theories of grammar as well as to construc-

tion grammar. Nevertheless, the central thesis of this chapter, that constructions

are the primitive units of syntactic representation, virtually requires that syntac-

tic theory be a variety of construction grammar. One consequence of the Radical

Construction Grammar position, however, is that the penultimate construct in

Table ., atomic schematic constructions (syntactic categories), does not exist in

the Radical Construction Grammar model (see §..).

1.3.2. Syntactic and semantic structure: the anatomy of a construction

In this section, I will introduce fundamental concepts and descriptive terms for

the analysis of the structure of a grammatical construction. The concepts in this

section form the basis of any syntactic theory, including Radical Construction

Grammar, although they are combined in different ways in different syntactic the-

ories. I will adhere to conventional terminology in construction grammar as much

as possible.

Grammatical constructions in construction grammar, like the lexicon in

other syntactic theories, consist of pairings of form and meaning that are at least

partially arbitrary. Even the most general syntactic constructions have corres-

ponding general rules of semantic interpretation. Thus, constructions are funda-

mentally units, as represented in Figure . (compare Langacker

: ).


Figure .. The symbolic structure of a construction


Figure .. The relation between form and function in a componential syntactic theory

Figure .. The relation between form and function in construction grammar

In the componential model, the various syntactic structures are organized inde-

pendently of the corresponding semantic structures, as represented by the bold

boxes in Figure .. In construction grammar, the basic linguistic units are

The term ‘meaning’ is intended to represent all of the

aspects of a construction’s function, which may include not only properties of

the situation described by the utterance but also properties of the discourse in

which the utterance is found (such as use of the Definite Article to indicate that

the object referred to is known to both speaker and hearer) and of the pragmatic

situation of the interlocutors (e.g. the use of a construction such as What a beau-

tiful cat! to convey the speaker’s surprise). In this book, I will use the terms

‘meaning’ and ‘semantic’ to refer to any conventionalized feature of a construc-

tion’s function.

The central essential difference between componential syntactic theories

and construction grammar is that the symbolic link between form and con-

ventional meaning is internal to a construction in the latter, but is external to the

syntactic and semantic components in the former (as linking rules). Figures .

and . compare construction grammar and a componential syntactic theory

on this parameter, highlighting in boldface the essential difference in the two

models.

The box notation used in Figure .b is simply a notational variant of the

bracket notation used in Figure .a (Langacker ; Kay and Fillmore ).

Thus, we can see that both the generative grammatical representation and the con-

struction grammar representation share the fundamental part-whole or -

structure of grammatical units: the sentence Heather sings is made up of

two parts, the Subject Heather and the Predicate sings.

The brackets in Figure .a are labeled with syntactic category labels, while the

corresponding boxes in the syntactic structure of Figure .b are not labeled. This

does not mean that the boxed structures in Figure .b are all of the same syn-

tactic type. Construction grammarians, of course, assume that syntactic units


Figure .. Simplified generative and construction grammar representations of Heather

sings

3 Other theories that share construction grammar’s basis in symbolic units are Head-driven Phrase

Structure Grammar (HPSG; Pollard and Sag , ), and Semiotic Grammar (McGregor ).

However, these theories are not explicitly construction-based, although HPSG and Fillmore and Kay’s

version of construction grammar have converged in many respects.

symbolic, and are organized as symbolic units (see §..), as represented by the

bold boxes in Figure ..3 As a consequence, the internal structure of the basic

(symbolic) units in construction grammar is more complex than that of basic

units in the componential model.

The internal structure of a construction is the morphosyntactic structure of

sentences that instantiate constructions. For example, a simple intransitive sen-

tence like Heather sings is an of the Intransitive construction. If we

compare a simplified representation of Heather sings in generative grammar to a

simplified representation of the same in construction grammar, we can see that

they are actually rather similar except that the construction grammar representa-

tion is symbolic (Figure .).

belong to a variety of different syntactic categories. The boxes have been left un-

labeled because the nature of those categories is precisely the issue to be addressed

in this chapter. All that it is necessary for us to assume at the moment is the mero-

nomic structure of the syntactic structure of a construction.

Beyond the meronomic structure of grammatical units, generative theories and

construction grammar diverge. First, as we have already noted, construction

grammar treats grammatical units as fundamentally symbolic, that is, pairings of

grammatical form and the corresponding meaning or . As

a consequence, the representation of a construction includes correspondence rela-

tions between the form and the meaning of the construction. We will call these

correspondence relations .

Since we will be talking about the formal or syntactic structure of a construc-

tion and also the semantic structure of a construction, it will be convenient to use

different names for the parts of a syntactic structure and the parts of a semantic

structure. We will call the parts of the syntactic structure and parts of

the semantic structure . Thus, a symbolic link joins an element of

the syntactic structure of a construction to a component of the semantic struc-

ture of that construction. There is also a symbolic link joining the whole syntac-

tic structure to the whole semantic structure (the middle symbolic link in Figure

.b). This symbolic link is the construction grammar representation of the

fact that the syntactic structure of the Intransitive construction symbolizes a

unary–valency predicate–argument semantic structure. Each element plus corres-

ponding component is a part of the whole construction (form + meaning) as

well. I will use the term to describe a symbolic part (element + component)

of a construction. That is, the construction as a symbolic whole is made up of

symbolic units as parts. The symbolic units of Heather sings are not indicated in

Figure .b for clarity’s sake; but all three types of parts of constructions are illus-

trated in Figure . (compare Langacker : , Fig. .a; Figure . suppresses

links between parts of the construction for clarity).


Figure .. Elements, components, and units of a construction

Figure .b has two other relations apart from the symbolic relation: one joining

the two syntactic elements and one joining the two semantic components. The

link joining the two semantic components describes a that

holds between the two components, in this case some sort of event–participant

relation. Thus, the semantic structure of a construction is assumed to be (poten-

tially) complex, made up of semantic components among which certain seman-

tic relations hold. This is all that we need to assume about the nature of semantic

structure in order to construct the arguments in this book, in particular, the argu-

ments in Part II. I believe that these assumptions are uncontroversial for most if

not all syntactic theorists.

The link joining the two syntactic elements in Figure .b is a -

. The syntactic relation does not obviously correspond directly to anything

in the generative grammar representation in Figure .. This is because the repre-

sentation of syntactic relations in most syntactic theories is more complex than a

simple syntactic link. In fact, we can identify three layers in the usual analysis of

syntactic relations:

the abstract syntactic relation;

the means of representing the abstract syntactic relation;

the overt manifestation of the abstract syntactic relation.

The first layer is the itself, such as the

Subject–Verb relation holding between Heather and sings in the construction

grammar representation in Figure .. This is intended to be a neutral way of char-

acterizing the syntactic structure of the construction.

The second layer is the the abstract syntactic relation.

Different syntactic theories use different means for representing abstract syntac-

tic relations. For example, generative grammar uses to represent

abstract syntactic relations. The labeled bracketing in Figure .a is a shorthand

for representing the constituency structure in ():

()

In other words, the generative grammatical representation in Figure .a displays

the means of representation of the syntactic relation Subject–Verb, not the

abstract syntactic relation itself.

Generative grammar uses constituency relations to represent many different

sorts of syntactic relations. For example, constituency is used to distinguish so-

called “grammatical relations” such as subject and object, as in Figure .. Other

theories, such as Word Grammar, represent a “grammatical relation” more directly

as a between the Verb and the Subject, as in Figure .. Still further

theories, such as Lexical-Functional Grammar, use a combination of constituency

S

NP VP

Heather sings


The third layer is the of the abstract syntactic relation,

that is, the linguistic evidence supporting the existence of a syntactic relation.

Examples of the overt manifestation of the syntactic relation between Heather and

sings are the contiguity of Heather and sings, the word order (Heather precedes

sings), and the overt expression of the suffix -s on the Verb that agrees with Heather

in Person and Number (rd Person Singular).

I will continue to represent only abstract syntactic relations in the construc-

tion diagrams in the remainder of this book. This is because in Part II, I will argue

that Radical Construction Grammar should dispense with abstract syntactic rela-

tions entirely. Thus, the means of representation of syntactic relations discussed

above are rendered vacuous. Instead, the putative overt manifestation of syntac-

tic relations exemplified above will be argued to be the manifestation of symbolic

relations.

One final theoretical/terminological point remains to be made. The analysis of

syntactic structure is unfortunately confounded by an ambiguity in much tradi-

tional syntax terminology. We can illustrate this with the example of the term

‘Subject’ in the Intransitive Clause construction in Figure ., illustrated once

again by the sentence Heather sings.


Figure .. Representation of subject and object in generative grammatical theories

Figure .. Dependency representation of syntactic relation between Heather and sings

Figure .. Roles in the Intransitive construction

and dependency to represent different types of syntactic relations holding between

elements of a syntactic structure.

The term ‘Subject’ can mean one of two things. It can describe the of a

particular element of the construction, that is, a - or rela-

tion between the element labeled ‘Subject’ in the Intransitive construction and the

Intransitive Construction as a whole. This is the sense in which one says that

Heather is the Subject I C Heather sings. This part-

whole relation is represented implicitly in Figure . by the nesting of the box for

Heather inside the box for the whole construction Heather sings.

The Subject role defines a grammatical category, namely those words (more

precisely, phrases) that can fill the Subject role. The Subject category is the cat-

egory of possible fillers of the Subject role in a Clause construction; some ex-

amples are given in ():

() a. Jennifer ran across the field.

b. Larry found $.

c. The car hit a tree.

The term ‘Subject’ can also describe a syntactic between one element of

the construction—the Subject—and another element of the construction—the

Verb. This is the sense in which one says that Heather is the Subject V

sings. As noted above, the Subject syntactic relation is assumed to be manifested

by a variety of properties, such as the case form of the noun phrase (in English,

with Pronouns), agreement of the Verb (in English with rd Person Subjects in

Present Tense), and word order (in English, preverbal position):

() a. She sings madrigals.

b. *Her sings madrigals.

c. *She sing madrigals.

d. *Madrigals sings she.

In other words, the term ‘Subject’ confounds two different types of relations in a

construction: the of the part in the whole, and the of one part to

another part (this terminological distinction is taken from Kay ). The differ-

ence between the two is illustrated in Figure ..


Figure .. The difference between syntactic roles and syntactic relations

In Part II of this book, I will argue against the existence of syntactic relations.

On the other hand, I assume the existence of syntactic roles in constructions in

Radical Construction Grammar. In fact, the part-whole structure of constructions

is the only sort of syntactic structure that Radical Construction Grammar calls

for. While this may appear to be an impoverished view of syntactic structure in

comparison to other syntactic theories, it must not be forgotten that Radical Con-

struction Grammar, like construction grammar in general, also posits symbolic

relations and semantic relations in a construction. It is in these latter structures

that significant cross-linguistic generalizations can be found.

1.3.3. The organization of constructions in a construction grammar

In Figure . (§..), I presented construction grammar as an inventory of con-

structions, in the broad sense of the latter term (including words, morphemes,

morphological structures, syntactic constructions, etc.). However, constructions

are not merely an unstructured list in construction grammar. Constructions form

a of a speaker’s knowledge of the conventions of their

language (Langacker : –). This structured inventory is usually represented

by construction grammarians in terms of a of constructions.

Each construction constitutes a in the taxonomic network of constructions.

Any construction with unique, idiosyncratic morphological, syntactic, lexical,

semantic, pragmatic, discourse-functional properties must be represented as

an independent node in the constructional network in order to capture a speaker’s

knowledge of their language. That is, any quirk of a construction is sufficient to

represent that construction as an independent node. For example, the substantive

idiom [S kick the bucket] must be represented as an independent node because

it is semantically idiosyncratic. The more schematic but verb-specific construc-

tion [S kick O] must also be represented as an independent node in order to

specify its argument linking pattern (or in older generative grammar terms, its

subcategorization frame). Finally, the wholly schematic construction [S V

O] is represented as an independent node because this is how construction

grammar represents the Transitive Clause that is described by phrase structure

rules in generative grammar, such as S Æ NP VP and VP Æ V NP.

Of course, kick the bucket has the same argument structure pattern as ordinary

transitive uses of kick, and ordinary transitive uses of kick follow the same argu-

ment structure pattern as any transitive verb phrase. Each construction is simply

an of the more construction(s) in the chain [kick the

bucket]—[kick O]—[V ]. Thus, these constructions can be represented

in a , as in Figure ..

However, grammatical constructions do not form a strict taxonomic hierarchy.

One of the simplifications in the hierarchy of constructions in Figure . is the

exclusion of Tense-Aspect-Mood-Negation marking, expressed by Auxiliaries and

Verbal suffixes. If those parts of an utterance are included, then any construction

in the hierarchy in Figure . has multiple parents. For example, the sentence

[I didn’t sleep] is an instantiation of both the Intransitive Clause construction and

the Negative construction, as illustrated in Figure ..


The sentence [I didn’t sleep] thus has in the taxonomy of

constructions to which it belongs. This is a consequence of each construction

being a specification of the grammatical structure of its daughter con-

struction(s). For example, the Negative construction only specifies the structure

associated with the Subject, Verb and Auxiliary; it does not specify anything about

a Verb’s Object (if it has one), and so there is no representation of the Object in

the Negation construction in Figure ..

A construction typically provides only a partial specification of the structure of

an utterance. For example, the Ditransitive construction [S DV O

O], as in He gave her a book, only specifies the predicate and the linkings to its

arguments. It does not specify the order of elements, which can be different in,

for example, the Cleft construction, as in It was a book that he gave her. Nor does

the Ditransitive construction specify the presence or position of other elements

in an utterance, such as Modal Auxiliaries or Negation, whether in a Declarative

Sentence (where they are preverbal; see (a) ) or an Interrogative Sentence (where

the Auxiliary precedes the subject; see (b) ):

() a. He won’t give her the book.

b. Wouldn’t he give her the book?

Hence, any specific utterance’s structure is specified by a number of distinct

schematic constructions. Conversely, a schematic construction abstracts away

from the unspecified structural aspects of the class of utterances it describes. The

model of construction grammar conforms to Langacker’s content requirement for


Figure .. A taxonomic hierarchy of clause types

Figure .. Multiple parents in a construction taxonomy

a grammar: the only grammatical entities that are posited in the theory are gram-

matical units and schematizations of those units (§.).

The taxonomic organization of syntactic knowledge in construction grammar

is simply the organization of syntactic structures into categories based on their

grammatical properties. As such, it might be expected that syntactic knowledge

follows the same principles of organization of other categories, in particular

lexical categories, since constructions are part of a representational continuum

with lexical items. This in fact appears to be true.

Some constructions are polysemous, with multiple senses or uses. An example

of a construction with multiple sense would be the different senses of the English

Present Perfect:

() a. President Clinton has visited Kosovo. [existential reading]

b. President Clinton has announced that America will invade Kosovo!

[“hot news” reading]

Some constructions have meanings that are metaphorical extensions from their

basic meaning, just as many words do. An example of a metaphorical extension

of a construction is the Perceptual Deictic There-construction, illustrated in (),

which is a metaphorical extension from the Central Deictic There-construction

illustrated in () (Lakoff : , ):

() a. Here comes the beep.

b. There’s the beep.

() There’s Harry.

The Perceptual Deictic describes the impending (a) or just-realized (b) acti-

vation of a nonvisual perceptual stimulus, for example an alarm clock that is about

to go off. To express this meaning, the Perceptual Deictic uses the metaphor of

deictic motion of a physical entity in physical space, expressed in the Presenta-

tional Deictic (Lakoff : ; see Goldberg : – for another example of a

metaphorical extension of a construction).

Constructions thus display many of the same properties of lexical items, such

as polysemy and metaphorical extensions. More generally, constructions (like

lexical items) represent , and construction gammar draws on cogni-

tive theories of categorization in its modeling of construction taxonomies.

For example, an issue that has attracted attention in the construction grammar

literature is the question: where is grammatical information represented in the

construction taxonomy? One school of thought argues that information should

not be redundantly represented in taxonomy. For example, the representation of

[kick the bucket] would only specify the idiosyncratic semantic interpretation of

that phrase; specification of the position and form of kick’s Direct Object would

then be from the more schematic superordinate constructions in

the taxonomy. This school of thought searches for the


construction schema (see §.). Kay and Fillmore’s () Construction Grammar

is a representative of this school of thought.

Another school of thought argues that grammatical information can be, and in

many cases should be, redundantly represented in the taxonomic hierarchy. Evi-

dence for redundant representation of grammatical information in the taxonomic

hierarchy comes from psycholinguistic evidence that patterns of frequency of use

determine the level of representation of grammatical knowledge in a speaker’s

mind. This school of thought, in other words, advocates a -

of grammatical representation (see e.g. Bybee ; Langacker ).

For example, a high token frequency of a particular word form or syntactic con-

struction will lead to the storage or of that word form or con-

struction even if its grammatical properties are predictable from taxonomically

superordinate constructions. Also, it is argued that , represented by

the entrenchment of a more abstract schema, is a function of its type frequency,

that is, the frequency of different instances of the schema. For example, the

productivity of the Past Tense schema [V-ed] is taken to mean that this schema

is highly entrenched in an English speaker’s mind, and that in turn is due to high

type frequency, that is, the large number of different Verbs that form their Past

Tense according to this schema. Conversely, low token frequency, that is, infre-

quent activation of the constructional schema in language use, can lead to decay

and loss of entrenchment of the construction schema over time. In other words,

the degree of generality of construction schemas, and the location of gram-

matical information in the taxonomic network, is an empirical question to be

answered by empirical studies of frequency patterns and psycholinguistic research

on entrenchment and productivity of schematic constructions.

A theory of grammar attempts to answer the question posed in §..: what is

the grammatical knowledge of a speaker? The usage-based model attempts to

answer that question. The development of usage-based models of grammatical

representation, and the psycholinguistic and linguistic evidence for them, is still

at an early stage. A considerable amount of research has been done on the usage-

based model in morphology, but the usage-based model has only very recently

been applied to syntactic constructions (see Cruse and Croft in press, chapter

for a survey of the literature). The usage-based model holds out the promise of a

psychologically principled way to determine which constructions of a language

should be independently represented as nodes in the taxonomic network, that is

which constructions are actually entrenched in a speaker’s mind.

There is another important principle of the organization of grammatical

knowledge, however, which is supported by evidence from typological universals

of language. A speaker’s knowledge of constructions is also organized by the rela-

tions between the meanings of the constructions. The semantic relationships

between constructional meanings can be represented in terms of a structured

. Conceptual space and the principles governing it will be

introduced in §., and the conceptual space model of grammatical organization


will play a significant part in the analyses proposed by Radical Construction

Grammar in the rest of this book.

1.4. Distributional Analysis and Cross-linguistic Universals

Up to this point, we have described distributional analysis as it applies to a single

language. However, distributional analysis is also a method that has been used for

uncovering Universal Grammar, that is, syntactic categories and roles that are

found across languages, not just within a single language. However, two serious

problems arise when distributional analysis is applied across languages. This

section describes these problems and two opposing approaches to resolving the

problems.

1.4.1. Problems in using distributional analysis across languages

1.4.1.1. Nonuniversality of constructions

The first problem is that the constructions used to define the categories in ques-

tion in one language (say, English) are missing in the language to be analyzed. For

example, inflectional criteria are commonly used to distinguish parts of speech

(noun, verb, and adjective) in languages, such that inflection for number, gender,

and case identify nouns and inflection for agreement, tense, aspect, and mood

identify verbs. However, Vietnamese lacks all morphological inflection (Emenean

: ). Hence inflection cannot be used to define parts of speech in Vietnamese.

Likewise, many of the criteria for Subject vs. Object given for English in ()–()

above are absent in Wardaman, an Australian aboriginal language. Wardaman

lacks infinitival complements and “Conjunction Reduction” coordination

(Merlan ), the constructions used in () and () to define Subject in English.

This problem is well-known. It was observed by the American structuralists

who codified distributional analysis. Before listing some typical inflectional cri-

teria for defining parts of speech across languages, Nida writes in a footnote,

‘There are so many languages in which the following generalizations do not hold

that one is tempted to avoid such statements altogether’ (Nida : , fn ).

This problem is well-known because it is common. Languages differ in the con-

structions that they possess. English is also lacking in constructions that are taken

to be relevant for defining parts of speech and syntactic roles in other languages.

For example, the Russian Noun category is defined by its inflection for Case as

well as Number, but English lacks case inflection.

Yet the distributional method by itself cannot solve this problem. The distri-

butional method does not tell us whether or not Vietnamese has parts of speech,

let alone the same parts of speech that English does. Likewise, the distributional

method does not tell us if the syntactic roles of Wardaman are the same as those

in English, since Wardaman lacks some of the constructions defining syntactic

roles in English.


1.4.1.2. Wildly different distributions across languages

A second serious problem is that even when the relevant constructions exist in

the language in question, they give wildly different distributions and hence wildly

different categories from those found in English and other familiar European lan-

guages. For example, Makah, a native American language of the Pacific North-

west, has inflections for Agreement, Aspect, and Mood, which are used as criteria

for the category Verb in English and other European languages. But virtually all

semantic classes of words can be inflected for Person-Aspect-Mood in Makah,

including what are English Verbs (), Nouns (), Adjectives (), and Adverbs

(; Jacobson : – see §..).

() k’upsil baʔas ʔu·yuq

point:MOM:IND: house

‘He’s pointing at the house.’

() baba�dis

white.man:IND:SG

‘I’m a white man.’

() ʔi·ʔi·x.wʔi

big:IND:

‘He’s big.’

() hu·ʔax.is haʔukw’ap

still:IND:SG eat:

‘I’m still feeding him.’

Again, the distributional method does not tell us whether or not the category

defined by Person-Aspect-Mood inflection in Makah is the same as the category

defined by Person-Tense-Mood inflection in English. Like the problem of missing

constructions, the problem of wildly different distribution is a cross-linguistically

common one. Languages differ considerably in the range of words that occur in

either nominal or verbal inflections. Likewise, languages differ considerably in the

range of noun phrases that can be “null” (i.e. not be present) in a clause, one of

the chief criteria for the category Subject in English and other languages. In War-

daman, for example, any Noun Phrase can be “null”, and so the criteria for Subject

role illustrated in ()–() cannot be applied to Wardaman, or rather, it would

make all Noun Phrases into “Subjects”.

1.4.2. Cross-linguistic methodological opportunism and its problems

The commonest solution to these problems is what I will call -

. Cross-linguistic methodological opportunism

uses language-specific criteria when the general criteria do not exist in the lan-

guage, or when the general criteria give the “wrong” results according to one’s

theory. For example, one can examine other constructions in Vietnamese and

Makah that differentiate word classes in a way that corresponds to the parts of

speech in European languages; or one can ignore the fact that some constructions


defining Subject in English are absent in Wardaman, and use the constructions

that do exist in Wardaman to define Subjects in that language.

But cross-linguistic methodological opportunism is just that: opportunistic. It

suffers from two interrelated and fatal problems. The first is that there is no a

priori way to decide which criteria (if any) are relevant to deciding that a particu-

lar category is an instantiation of a universal category like Noun or Subject across

languages. One might propose that inflection for agreement and tense-mood-

aspect will be the criterion for the category Verb across languages. But why? No

reason has been given to do so. And if one does so, then one will have to conclude

that all words are Verbs in Makah and no words are Verbs in Vietnamese, which

is hardly a savory conclusion for a theory that posits Verbs as a part of Universal

Grammar.

The second problem is that the choice of criteria looks suspiciously like serving

a priori theoretical assumptions. For example, there has been an ongoing debate

about whether native American languages of the Pacific Northwest, including

Makah, have the Noun–Verb distinction. Those who argue against the Noun–

Verb distinction take the distributional pattern in ()–(), that is, inflection

in predication, as criterial. Those who argue in favor of the Noun–Verb dis-

tinction, such as Jacobsen, use distributional patterns in other constructions to

differentiate the word classes. Without prior agreement or some principled means

for specifying which constructions define a category across languages, analysts can

use whatever constructions they wish in order to come to whatever conclusions

they wish.

Cross-linguistic methodological opportunism in identifying categories across

languages is unprincipled and ad hoc. In other words, cross-linguistic method-

ological opportunism is not a rigorous scientific method for discovering the prop-

erties of Universal Grammar. Yet it is a very widely used form of cross-linguistic

argumentation.

For example, Jelinek and Demers argue that Straits Salish does not distinguish

nouns, verbs, and adjectives, because all can appear in the Predication

construction ( (a–c), with the enclitics =l =sxw ‘==.’), and in the

Determination construction ( (a–c), with the Article c ; Jelinek and Demers

: –):

() a. t’iləm=l =sxw ‘you sang’

b. si’em=l =sxw ‘you were a chief ’

c. sey’si’=l =sxw ‘you were afraid’

() a. c t’iləm=lə ‘the (one who) sang’

b. c si’em=lə ‘the (one who) was a chief ’

c. c sey’si’=lə ‘the (one who) was afraid’

But van Eijk and Hess () observe the same distributional facts for the closely

related languages Lillooet and Lushootseed, and yet conclude that Lillooet and

Lushootseed do distinguish Noun and Verb. Their argument is based on the

əə


distribution pattern of Possessive affixes and Aspect inflection, which divide

Lillooet and Lushootseed roots into Noun and Verb categories in their analysis

(van Eijk and Hess : –).

Jelinek and Demers note that in Straits Salish also, only a subset of roots may

take Possessive affixes (Jelinek and Demers : ). But they point out that

“Nouns” may occur in the Predication construction (ibid.; see (b) ), and then

argue that the fact that “Nouns” with Possessive affixes may occur in the Predica-

tion and Determination constructions is the deciding case: ‘they have the same

syntax as any other predicate’ (ibid.: ).

Here we see a common pattern in syntactic debate. The distributional facts

show that there are some similarities and some differences between two gram-

matical phenomena. One set of analysts (e.g. Jelinek and Demers) takes a

“lumping” approach, arguing that certain distributional differences are superficial

compared to the underlying grammatical unity. The other analysts (e.g. van Eijk

and Hess) take a “splitting” approach, arguing that the distributional differences

really are significant and require a distinct analysis for the two phenomena. Yet

there is no a priori way to resolve the question: the “lumper” overlooks the mis-

matches in distribution, and the “splitter” overlooks the generalizations. Another

way must be sought out of this dilemma.

1.4.3. An alternative view: there is no universal inventory of atomic primitives

Dryer (b) proposes a different solution to the problems posed in §.., at least

for functionalist approaches to syntax (I will discuss formal approaches to syntax

below). Dryer bases his argument on the problems posed by identifying gram-

matical relations such as Subject and Object across languages, but he invokes par-

allels with phonemes and parts of speech, and I will briefly present Dryer’s

argument in generalized form.

Dryer suggests that the following four things might be proposed to exist in the

domain of grammar (adapted from Dryer b: –):

() a. categories and relations in particular languages

b. similarities among these language-particular categories and relations

c. functional, cognitive and semantic explanations for these similarities

d. categories and relations in a cross-linguistic sense

Syntactic theorists, including many functionalist theorists, assume the existence

of (d), that is, categories and relations. These universal categories

and relations are then instantiated in the grammars of particular languages. In

other words, (a) is just an instantiation of (d). In this view, (b)—the fact

that categories across languages are similar, not identical—is due to language-

particular peculiarities that do not affect the overall architecture of Universal

Grammar.

But in fact there is a wide range of cross-linguistic variation in syntactic cat-

egories and roles and other basic syntactic phenomena, as we have already seen,


and will see in later chapters. Dryer argues that a functionalist syntactician should

accept the uniqueness of language-particular grammatical relations, namely

accept (a) and reject (d). Dryer argues that such an approach is palatable to

a functionalist approach to language because there similarities among

language-particular grammatical relations, and the functionalist offers functional,

cognitive, and semantic explanations for these similarities, (c). That is, a func-

tionalist can offer an explanatory theory of language without positing universal

grammatical relations. Dryer writes:

the search for an understanding of the similarities and differences among grammatical rela-

tions in different languages will be impeded if we make the mistake of thinking of gram-

matical relations as crosslinguistic categories, and will be more successful if we bear in mind

that grammatical relations are unique to every language (Dryer b: ).

the similarities among languages [with respect to grammatical relations] can be explained

directly in terms of functional and cognitive principles that underlie language and which

cause languages to be the way they are (Dryer b: ).

Dryer suggests that formal theories of syntax can posit universal grammatical

relations without a problem: ‘The view that grammatical relations are cross-

linguistic notions makes sense if one adopts the view of formal linguistics that

explanation of language is largely internal to language’ (Dryer b: ).

However, Dryer does not elaborate this point, and it is not clear why it should be

true. Formalist linguists face the same methodological problems with distribu-

tional analysis, and the same facts of cross-linguistic variation, as functionalist lin-

guists do. If the formalist explanation is largely internal to a language, which is

essentially correct, then there are no criteria for identifying categories as the same

across languages.

In fact, the argument against universal categories and relations is not functional

at all. The problems with applying the distributional method across languages

involve constructions and the distributional patterns they define, not any particu-

lar functionalist analysis of the constructions and their distributional patterns.

The argument against universal categories and relations is a fundamentally -

one. Moreover, the empirical argument against universal categories is not

based on an esoteric fact of a single little-known language; it is based on the well-

known and pervasive grammatical diversity of languages.

The alternative view, rejecting universal categories and relations, has a number

of advantages over the standard view. The alternative view avoids the inconsis-

tencies of cross-linguistic methodological opportunism, which leads to undecid-

able disputes over questions such as whether Makah has the Noun–Verb

distinction or whether Vietnamese can be said to have parts of speech at all. There

is no need to ignore distributional patterns that do not fit the assumed universal

categories. The alternative view allows each language to be itself: it respects the

grammatical diversity of languages, and the uniqueness of each language’s

grammar.


In this respect, the alternative view is little different from the view espoused

by American structuralists: ‘no logical scheme of the parts of speech—their

number, nature and necessary confines—is of the slightest interest to the linguist.

Each language has its own scheme’ (Sapir : ). Instead, the alternative

view allows the analyst to focus on the similarities and differences across lan-

guages. Positing universal categories would imply identical behavior across

languages, which is empirically false. The more interesting and significant task

(pace Sapir) is to explain similarities and differences of categories and relations

across languages.

Of course, abandoning universal categories and relations leads to a very differ-

ent view of Universal Grammar. Under the alternative view, Universal Grammar

does not consist of an inventory of universal categories and relations available to

all speakers. For a functionalist, this is not a serious problem, since a functional-

ist seeks universals of language in cognition and discourse. This does not mean,

of course, that a functionalist can ignore the question of what the structure of a

grammar looks like under the alternative view. It just means that the formal struc-

tures in grammars are language-particular, and universals of language must be

sought elsewhere.

For a formalist, accepting the alternative view poses a more serious challenge.

Universal Grammar in a formal sense could not consist of anything more than

very general constraints on types of categories and their configuration in syntac-

tic structures or constructions. In principle, that appears to be the direction that

Chomskyan generative grammar has headed: general constraints on syntactic

structure but a proliferation of syntactic categories. In practice, however, the syn-

tactic categories are assumed to be cross-linguistically universal, and the same cat-

egories (or a subset thereof) are posited for every language. This practice also

holds for other formal syntactic theories.

Given the actual practice of formal syntacticians, then, the alternative view

argued for here is very serious, and one might wish to find ways to accommodate

cross-linguistic variation without giving up on universal categories and relations.

I do not see how one can do so without falling back into cross-linguistic method-

ological opportunism. However, the theoretical quandary is in fact more serious,

because the same problems arise in the analysis of a single language.

1.5. Distributional Analysis and the Representation of

Particular Language Grammars

1.5.1. Problems in using distributional analysis in particular languages

1.5.1.1. Mismatch in distribution between constructions

The fundamental problem with distributional analysis across languages is the

difference or in distribution patterns for the same criterion or con-

struction across languages. But within a single language, the different criteria


(constructions) used to define a particular category often have different distribu-

tional patterns as well. That is, there are mismatches in the distribution patterns

defined by different constructions in a single language.

A simple example of mismatch in distribution is found with the arguments for

distinguishing English Direct Objects from Obliques. English Direct Objects can

occur as the Noun Phrase that immediately follows the Verb in the Active Voice

and lacks a Preposition, as in (a):

() a. Jack kissed Janet.

b. Janet was kissed by Jack.

Another test or criterion for Direct Object status in English is occurrence as the

Subject of the Verb in the counterpart Passive Voice, as illustrated in (b).

Obliques contrast with Direct Objects on both counts. Obliques cannot occur

without a Preposition (compare (a) to (b) ), and cannot occur as Passive Sub-

jects (c):

() a. *The old man walked a cane.

b. The old man walked with a cane.

c. *A cane was walked with by the old man.

The distribution pattern for Direct Objects vs. Obliques is given in Table ..


Table .. Distribution of arguments across tests for Direct

Object status

[S V __] [__ be V:P by O]

Direct Object: ÷ ÷Oblique: * *

Table . makes it appear that the two tests for Direct Object status coincide.

However, that is not the case. There are postverbal prepositionless Noun Phrases

that cannot occur as Passive Subjects:

() a. Jack weighs pounds.

b. * pounds is weighed by Jack.

() a. witnessed the demise of years of Tory rule in Britain.

b. *The demise of years of Tory rule in Britain was witnessed by .

And there are Oblique Objects of Prepositions that can occur as Passive Subjects:

() a. Claude Debussy lived in this house.

b. *Claude Debussy lived this house.

c. This house was lived in by Claude Debussy.

Mismatches in distributional patterns are the norm in languages. The Ameri-

can structuralists were aware of this fact. Moreover, the more constructions that

one uses to define categories, the larger the number of distinct categories that

would be observed, and the smaller each of those categories would be. In fact, if

one takes all distributional criteria seriously, one would end up with a very large

number of syntactic categories and relations, each of which would have very few

members. Again, the American structuralists were aware of this problem:

Form-classes are not mutually exclusive, but cross each other and overlap and are included

one within the other, and so on (Bloomfield : ).

in many cases the complete adherence to morpheme-distribution classes would lead to a

relatively large number of different classes (Harris : ).

If we seek to form classes of morphemes such that all the morphemes in a particular class

will have identical distributions, we will frequently achieve little success. It will often be

found that few morphemes occur in precisely all the environments in which some other

morphemes occur, and in no other environments (Harris : , discussing analysis of

a corpus).

This is not simply a matter of speculative extrapolation: this conclusion has also

been empirically verified in at least one large-scale formal grammatical model. In

a very large grammar of French developed by Maurice Gross and colleagues, con-

taining rules covering , lexical items, no two lexical items had exactly

the same distribution, and no two rules had exactly the same domain of applica-

tion (Gross : –).

Again, distributional analysis does not provide an answer to the problem as to

which construction, or both, or neither should be used to define the category of

English Direct Objects, or any other category for that matter.

1.5.1.2. Subclasses and multiple class membership

Some cases of distributional mismatches have been analyzed as instances of syn-

tactic subclasses rather than as distinct syntactic classes. For example, the English


Table .. Distributional patterns for Active Object and

Passive Subject in English

Example [S V __] [__ be V:P by O]

÷ ÷ * *

, ÷ *

* ÷

In other words, all possible distributional patterns for these two constructions

are attested, as shown in Table ..

Noun category, which can be defined by occurrence as the Head of a Noun Phrase

(see (a–b) ), is traditionally divided into two subclasses, Count and Mass, in

part by the occurrence of the former but not the latter with the Plural (see

(a–b) ):

() a. the student/the book/etc.

b. the mud/the air/etc.

() a. student-s/book-s/etc.

b. *mud-s/*air-s/etc.

The distributional pattern is given in Table .; one construction defines the class,

and the other construction differentiates the subclasses.


Table .. Subclasses of one word class?

[__-s] [the __]NP

Count Noun: student, book, etc. ÷ ÷Mass Noun: mud, air, etc. * ÷

Table .. Separate word classes?

[the __]NP [the __ N]NP

Noun: student, book, etc. ÷ ÷Adjective: tall, sad, etc. * ÷

On the other hand, English Nouns and Adjectives do occur in some of the same

constructions: they both take the Copula be when predicated (a–b), and they

can both occur as prenominal Modifiers (a–b):

() a. She is a student/This is a book/etc.

b. She is tall/She is sad/etc.

() a. student discount/book warehouse/etc.

b. tall girl/sad woman/etc.

English Nouns and Adjectives can also be differentiated: for example, Nouns can

be the Heads of Noun Phrases while Adjectives cannot:

() a. the student/the book/etc.

b. *the tall/*the sad/etc.

The distribution pattern of () vs. () is given in Table ..

But the distributional patterns in Tables . and . are the same. In other words,

there is no a priori basis for deciding which constructions provide sufficient

conditions for separate word classes and which constructions merely define sub-

classes of one word class. The distributional method alone cannot decide this

question. As Schachter puts it:

It must be acknowledged, however, that there is not always a clear basis for deciding

whether two distinguishable open classes of words that occur in a language should be iden-

tified as different parts of speech or as subclasses of a single part of speech. . . . What this

means is that there may in some cases be considerable arbitrariness in the identification of

two open word classes as distinct parts of speech rather than subclasses of a single part of

speech (Schachter : –).

The same is true of another strategy for dealing with mismatches in distribution,

namely allowing for multiple class membership. For instance, another construc-

tion distinguishing Adjectives from Nouns is occurrence with the Anaphoric Head

one:

() a. *the box one/*the woman one/etc.

b. the tall one/the sad one/etc.

c. the rich one/the poor one/etc.

However, the words in (c) can also occur as the Head of a Noun Phrase

themselves:

() a. the box/the woman/etc.

b. *the tall/*the sad/etc.

c. the rich/the poor/etc.

Words like rich/poor are usually analyzed as being both Nouns and Adjectives, that

is, as having multiple class membership. The distribution pattern defined by the

constructions in ()–() is given in Table .; the first construction defines

Nouns and the second defines Adjectives.


Table .. Multiple word class membership?

[the __]NP [the __ one]NP

Noun: box, woman, etc. ÷ *

Adjective: tall, sad, etc. * ÷Noun and Adjective: ÷ ÷

rich, poor, etc.

On the other hand, the Genitive Noun Phrase with the Possessive enclitic -’s is

usually treated as belonging to a different category from the Attributive Posses-

sive Pronouns and the Pronominal Possessive Pronouns. Distributionally, the

Genitive Noun Phrase can function attributively, that is, occur as a prenominal

Modifier, like the Attributive Possessive Pronouns:

() a. my book/your book/etc.

b. *mine book/*yours book/etc.

c. John’s book/Sally’s book/etc.

And the Genitive Noun Phrase can function pronominally, that is, occur as a Noun

Phrase on its own:

() a. *bigger than my/*bigger than your/etc.

b. bigger than mine/bigger than yours/etc.

c. bigger than John’s/bigger than Sally’s/etc.

The distribution pattern for ()–() is given in Table ..


Table .. A third word class?

[__ N] [bigger than __]

Attributive Possessive: my, your, etc. ÷ *

Pronominal Possessive: mine, yours, etc. * ÷Genitive Noun Phrase: John’s, Sally’s, etc. ÷ ÷

As with Tables .–., the distribution patterns in Tables .–. are the same.

In other words, there is no a priori basis for deciding which constructions are only

necessary conditions for word class membership (hence permit a third word class

analysis) and which constructions are sufficient to define separate word classes

(hence multiple class membership for words occurring in both constructions).

Again, the distributional method cannot decide this question.

1.5.1.3. Lack of exclusive partitioning of lexical items

A third problem with the distributional method is that there is often a lack of

exclusive partitioning of lexical items. For example, some English Nouns are

clearly Mass Nouns, as defined by their occurrence in the Bare Singular con-

struction without an Article (see () ):

() a. There’s mud on your boots.

b. *I found two muds on the carpet.

Other nouns are both Mass Nouns and Count Nouns, as defined by occurrence

with Numerals (compare the (a) and (b) sentences in ()–() ):

() a. There’s chocolate on your hands.

b. I ate only five chocolates.

() a. There’s hair on the sofa.

b. There’s a hair on the sofa.

The lack of exclusive partitioning of the lexical items in ()–() could be

accounted for by multiple class membership or by a third class. However,

it appears that almost any Count Noun can potentially occur in the Bare

Singular construction, as attested examples such as (b) demonstrate (see also

§..):

() a. The Walkers own three cars.

b. “There was a huge Buick there; just acres of car.”

[overheard by Mary Ellen Ryder, Manchester, April ]

Another example of the lack of exclusive partitioning of lexical items are the words

variously described as Adverbs, Prepositions and/or Particles (e.g. Biber et al. ,

§§.., ..–). I illustrate the problems with one word, down:

() a. She walked down (from the mountain).

b. She sat down.

c. She looked down (at the people below).

() a. She walked down the road.

b. *She sat down the table. [‘down at the table’]

c. She is sitting down the table from me.

d. *She looked down the people below.

e. She looked down the stairwell.

() a. The proposal went down badly.

b. After she left, he broke down.

c. He broke down the problem into five parts.

d. *He broke down it into five parts.

The examples in () are Adverb-like, indicating direction. However, down may

sometimes act like a spatial Preposition, as in (a,c,e)—but not (b,d). The

examples in () are usually called Particle uses of down. However, the examples

in (a–b) are more Adverb-like, while the example in (c) is more Object-like—

except when a pronominal object is involved (d). Down is of course only one

of very many words that behave slightly differently, yielding either many sub-

classes or many cases of criss-crossing multiple class membership, without any

clear picture of how many classes to posit, and with some indeterminacy in some

contexts. For example, regarding the attested examples (a–b), Biber et al. note

that in in (a) is more like a Particle because it can be omitted; but omission of

in in (b) is ‘less freely admissible and therefore behaves more like a preposition’

(Biber et al. : ):

() a. I knew that there was a man in there. ()

b. In there I feel skinnier than ever. ()

This problem is of a somewhat different type from the other problems described

with the distributional method. Indeterminacy of classification implies that distri-

butional patterns may be more flexible and fluid than is usually assumed, and

hence word classes are not as sharply defined as is commonly thought.


1.5.2. Language-internal methodological opportunism and its problems

The common solution to the problems posed by the distributional method in a

single language is analogous to the common solution to the cross-linguistic

problem. I will call it - .

Language-internal methodological opportunism simply selects a subset of

language-specific criteria to define a category when the criteria do not all match.

That subset of criteria, or possibly just one criterion, defines the category in ques-

tion. Mismatching distributions are ignored, or are used to define subclasses or

multiple class membership.

Language-internal methodological opportunism was proposed by Bloomfield

and Harris as a solution to the problems posed in §.., as can be seen by the con-

tinuation of the quotations from §...:

For this reason a system of parts of speech in a language like English cannot be set up in

any fully satisfactory way: our list of parts of speech will depend upon which functions we

take to be the most important (Bloomfield : ).

This means that we change over from correlating each morpheme with all its environments,

to correlating selected environments (frames) with all the morphemes that enter them

(Harris : ).

For example, the usual solution to the mismatch in the two criteria commonly

used to define Direct Objects in English is to select the criterion of Passive Subject,

or passivizability in the transformational metaphor, as the defining criterion of

Direct Object. Thus, pounds in (a) and the demise of years of Tory rule in

Britain in (a) are not Direct Objects in this approach.

Language-internal methodological opportunism is just as unsatisfactory as

its cross-linguistic cousin. There is no a priori way to decide which of several

constructions with mismatching distributions, or which subset of construc-

tions, should be chosen as criteria for identifying the category in question.

Why should passivizability be the criterion for defining the Direct Object

category? Why shouldn’t the criterion be occurrence as the postverbal pre-

positionless Noun Phrase in the Active construction? The choice of criteria

again looks suspiciously like serving a priori theoretical assumptions of the

analyst, for example a priori assumptions about what should or should not be

a Direct Object. Moreover, if one does choose one construction (or subset of

constructions) to define a category, then one still has not accounted for the

anomalous distribution pattern of the constructions that have been left out (in

this case, occurrence as the postverbal prepositionless Noun Phrase in the Active

construction).

Language-internal methodological opportunism, like cross-linguistic method-

ological opportunism, is unprincipled and ad hoc, and hence is not a rigorous

scientific method for discovering the properties of the grammar of a language.


Yet language-internal methodological opportunism is a widely used form of

argumentation in the analysis of particular language grammars.

I will give one illustrative example, from the debate between Larson and Jack-

endoff on the analysis of the Ditransitive (Double Object) construction in English

(Larson ; Jackendoff b; Larson ). Of course, a proper explication of

the arguments and counterarguments in these lengthy articles would take up more

space than can be afforded here. Much of the debate between Larson and Jack-

endoff does not involve a disagreement with the distributional facts that they

present, but instead turns on what theoretical construct in generative grammar is

preferable to analyze the distributional facts, and on semantic arguments. Here

we will restrict ourselves to two essential syntactic arguments presented in these

articles.

Larson analyzes the Indirect Object construction in (a) as a result of Verb

movement in the structure in (b) (Larson : –):

() a. send a letter to Mary

b.

Larson analyzes the Double Object construction in (a) as a result of the NP

movement in the structure in (b) (Jackendoff b: ):

() a. send Mary a letter

b.


In (b), the NP Mary is moved upwards and the NP a letter is “demoted” in com-

parison to its position in (b).

Larson further argues that the syntactic operation in () is essentially

identical to the Passive operation, illustrated in () (from Haegeman :

):

() a. This story is believed by the villagers.

b.

That is, both Passive and Dative Shift involve moving one NP upwards in the

tree—recipient in Dative Shift, patient in Passive—and “demoting” one NP—

theme in Dative Shift, agent in Passive—in comparison to the “demoted” NP’s

position in each construction’s counterpart—Indirect Object for Dative Shift,

Active for Passive.

Larson’s justification for the analysis in (b) is a variety of parallel distribu-

tional facts, such as the each . . . the other construction (Larson : , ; see

Barss and Lasnik ):

() a. Dative Shift: *I showed the other’s friend each man. cf.:

b. I showed each man the other’s socks.

() a. Passive: *The other boy was recommended by each mother. cf.:

b. Each mother recommended the other boy.

Larson notes differences between Dative Shift and Passive: for example, the latter

occurs in the [V-en] morphological construction (in our terms) but the former

does not:

() *given Mary a letter

However, he argues that other generative principles can account for the ‘apparent

differences’ (Larson : ).

Jackendoff argues for the traditional analysis of the Double Object construc-

tion as having two Object complements, as in (b) (Jackendoff b: ), and

argues that the distributional facts presented by Larson are in fact due to linear

order, not the nested constituent structure given in (b).


() a. show John himself

b.

For Jackendoff the difference in distribution between Dative Shift and Passive

illustrated in example () is not a mere ‘apparent difference’, but an argument

against Larson’s analysis.

Jackendoff further notes another distributional difference between Passive and

Dative Shift (Larson : and Jackendoff b: ):

() a. Dative Shift : I showed Mary herself.

b. Passive: *Bill was hit by himself.

Reflexivization of the “demoted” NP when it is coreferential with the “moved” NP

is grammatical in Dative Shift, but not in Passive. In his response to Jackendoff,

Larson does not mention this distributional fact, which suggests that this distri-

butional difference is not deemed sufficiently serious for him to abandon his

analysis.

Jackendoff also presents distributional evidence that the facts in ()–() are

also shared by certain other constructions, including the Locative Alternation con-

structions in () (see Jackendoff b: –):

() a. I loaded each set of books into the other’s box.

b. *I loaded the other’s box with each set of books.

Jackendoff argues that the presence of the Prepositions and the choice of Prepo-

sition in () poses a problem for Larson’s analysis (Jackendoff b: –). That

is, Jackendoff considers the distributional differences—presence of Prepositions

in the Locative Alternation construction in () vs. absence of Prepositions in

Dative Shift—significant enough to reject Larson’s analysis. In his response,

Larson accepts these distributional facts. However, Larson argues that ‘this does

not represent a “sharp” difference between dative and spray-load pairs’ (Larson

: ). Larson then points out still other distributional differences that under-

mine Jackendoff ’s analysis which I will not pursue here.

This brief sketch of two central arguments in these articles illustrates the same

common pattern in syntactic debate that was noted in §... The distributional

facts show that there are some similarities and some differences between two

grammatical phenomena. One analyst (Larson) takes a “lumping” approach and

the other analyst (Jackendoff) takes a “splitting” approach. Yet there is no a priori

way to resolve the question: the “lumper” overlooks the mismatches in distribu-

tion, and the “splitter” overlooks the generalizations. Again, another way must be

sought out of this dilemma.


1.5.3. An alternative view: there are no atomic grammatical primitives

In this section, I argue that the same solution offered by Dryer for cross-

linguistic categories should be applied to categories within a language. The fol-

lowing four things might be proposed to exist in a particular language grammar:

() a. categories and relations defined by particular constructions

b. similarities among these construction-specific categories and relations

c. functional, cognitive, and semantic explanations for these similarities

d. categories and relations in a cross-constructional sense

All of the syntactic theories referred to in §. assume the existence of (d), which

I will call categories and relations. These global categories and relations

are then instantiated in particular constructions in a particular language. In other

words, (a) is just an instantiation of (d). In this view, the fact that categories

across constructions are similar, not identical, is due to construction-specific

peculiarities that do not affect the overall architecture of the language’s grammar.

But in fact distributional criteria in general do not match, within or across lan-

guages. Yet there is no justification for deciding which distributional criteria are

the “right” ones for establishing syntactic primitives—categories and relations—

demanded by contemporary syntactic theories. These two basic facts call into

question the assumption of the existence of (d)—global categories and rela-

tions—by contemporary syntactic theories.

The real problem is in a logical inconsistency in the way the distributional

method is used, given the fact of distributional mismatches among constructions.

The distributional method is combined with the theoretical assumption that the

categories/relations defined by constructions are the syntactic primitives used to

represent grammatical knowledge. C -

—this is the distributional method. B

—this is the syntactic model of representation. T

. It would not be circular if there were no distributional

mismatches, because the facts of language would then conform to the theoretical

assumption. But the facts of language are otherwise.

Given the facts of language and the circularity of the standard approach to those

facts, one must discard either the distributional method or the assumption that

categories/relations are the theoretical primitives of syntactic theory. But how

can one discard the distributional method? What is the alternative? Yet cross-

linguistic and language-internal methodological opportunism essentially do

abandon the distributional method, at the price of empirical coverage and theor-

etical consistency. Methodological opportunism selects distributional tests at the

whim of the analyst, and ignores the evidence from other distributional tests that

do not match the analyst’s expectations, or else treats them as superficial or

peripheral.

Instead, I propose that we discard the assumption that syntactic structures are


made up of atomic primitives (language-universal or language-particular). C-

, , ,

. The categories and relations found in construc-

tions are derivative—just as the distributional method implies. This is Radical

Construction Grammar.

So-called theories of syntactic categories, grammatical relations (and also of

dependency, constituency, heads, argument vs. adjunct, main clause vs. subordi-

nate clause, etc.) are at best theories of the constructions used by the analyst to

argue for the existence of the categories, relations, etc. For example, if one takes

passivizability as the criterion for Direct Object in English, then one’s conclusions

will tell us something about the Passive, not about some allegedly global category

Direct Object. At worst, theories of categories, etc. are theories of nothing at all,

if the analyst does not apply his/her constructional tests consistently.

Such theories fail because of mismatches across languages and especially within

a language between the distributions defined by the constructions that are used

by the analyst. The mismatches are a theoretical problem because the assumption

of atomic syntactic primitives means that all the distributions should match up.

That is, not just similarity but identity of distribution is expected. Again, this argu-

ment against existing formal theories of syntax is not functional at all, and hence

is not dependent on a belief in ( c), that is, functional, cognitive, or semantic

explanations for similarities of categories across constructions. The argument is

simply empirical, based on the well-known and pervasive diversity of distribu-

tions defined by grammatical constructions.

In Radical Construction Grammar, the grammatical knowledge of a speaker

is knowledge of constructions (as form–meaning pairings), words (also as

form–meaning pairings), and the mapping between words and the constructions

they fit in. The mapping between words and the constructions is many-to-many:

a word fits into many different constructions, and constructional roles can be

filled by many different words. This knowledge is represented by the distribution

tables given in this chapter, though we will see in later chapters that there is more

structure to grammatical knowledge than the distributional tables give. This

model can of course be extended to constructions that have other constructions,

not just words, as component elements.

In Radical Construction Grammar, syntactic categories are derivative of—in

fact epiphenomenal to—the representation of grammatical knowledge. Syntactic

categories can be defined in two different ways. Categories can be defined

construction-specifically, as the class of fillers of a particular role in a single con-

struction. This definition corresponds to the columns in the distribution tables

given in this chapter. Categories can also be defined cross-constructionally, as the

class of fillers that has an identical distribution across the relevant roles for all

constructions of the language, or at least some specified set of constructions in

the language. This definition corresponds to the rows in the tables.

Either definition for syntactic categories can be analytically useful in some cir-


cumstances. But what is basic are the constructions, and the constructions define

the categories, either individually or jointly. Among other things, this approach

solves the problem of the lack of exclusive partitioning of lexical items into atomic

primitive categories: this fact reflects the more-than-occasional variability and

instability of the word–construction relationship, and the priority of construc-

tions in defining categories.

1.6. Radical Construction Grammar: Frequently Asked Questions

The fundamental thesis of Radical Construction Grammar, that constructions

and not categories are the primitive units of syntactic representation, is an idea

that is difficult to grasp because the opposite view is so deeply entrenched in lin-

guistic theory (indeed, it took me almost ten years to accept this point of view).

As a result, certain questions have been repeatedly asked in my presentations of

Radical Construction Grammar. This section will offer answers to these frequently

asked questions.

1.6.1. How can you have a syntactic theory without atomic primitive units?

The main conceptual hurdle to considering Radical Construction Grammar as a

logically possible theory of grammar is the assumption that the primitives of syn-

tactic theory must be atomic. But ‘atomic’ and ‘primitive’ are logically indepen-

dent concepts. A units are those that cannot be broken down into smaller

parts in the theory. P units are those whose structure and behavior

cannot be defined in terms of other units in the theory. Primitive elements need

not be atomic. The notions ‘atomic’ and ‘primitive’ can be dissociated.

The logical independence of ‘atomic’ and ‘primitive’ allows us to give a precise

definition of reductionist and nonreductionist scientific theories. Theories in

which the primitive theoretical constructs are atomic are theories.

A reductionist theory begins with the smallest units and defines the larger or more

complex units in terms of combinations of atomic primitive units. All of the

theories of syntactic representation referred to in §. are reductionist. Complex

syntactic structures, that is, constructions, are defined by—built up from—their

ultimately atomic primitive parts: syntactic categories and syntactic relations (such

as constituency and dependency). Contemporary syntactic theories differ chiefly

in the inventory of syntactic primitives and the rules governing their combination.

Theories in which the primitive theoretical constructs are complex are -

theories. A nonreductionist theory begins with the largest units

and defines the smaller ones in terms of their relation to the larger units. The par-

adigm example of a nonreductionist theory is the theory of perception proposed

by Gestalt psychology (Koffka ; Köhler ; Wertheimer ). In Gestalt

psychology, evidence is presented to the effect that the perception of features of

objects is influenced by the perceptual whole in which the feature is found.

Radical Construction Grammar is a nonreductionist theory of syntactic


representation. Constructions, not categories and relations, are the basic, primi-

tive units of syntactic representation. The categories and relations internal to con-

structions are derived from them, as described in §...

A nonreductionist theory does not deny that constructions (syntactic struc-

tures) are made up of parts. A nonreductionist theory of grammar, like reduc-

tionist theories, assumes that constructions, or more precisely, actual instances of

constructions, can be segmented into their constituent words and morphemes. A

nonreductionist theory differs from a reductionist theory in that it hypothesizes

that the whole is greater than the sum of its parts. The parts take their signifi-

cance—that is, are categorized—by virtue of the role that they play in the con-

struction as a whole.

The parts qua instances of a syntactic category do not have an independent

existence outside of the whole construction (or constructions) in which they play

a role. That is, syntactic categories are defined in terms of the construction(s) in

which they occur. A nonreductionist construction grammar therefore does not

have any atomic schematic constructions (see Table . in §..).

1.6.2. Can’t these facts be captured by a feature-based approach to categories, or

by a categorial grammar approach?

In some contemporary syntactic theories, certain syntactic categories are not

atomic but are instead decomposed into features. For example, there is a widely

cited decomposition of the major parts of speech into the features [±N, ±V]

(Chomsky ; Jackendoff ). In other syntactic theories, such as Head-driven

Phrase Structure Grammar and Fillmore and Kay’s Construction Grammar, the

categories themselves are features, e.g. <cat v> for Verb; but the category feature

is just one feature among many others that play a role in accounting for distrib-

utional patterns.

The use of features instead of categories may or may not be in the spirit of

Radical Construction Grammar. The use of features illustrated by [±N, ±V] rep-

resents an even greater degree of reductionism than that found in theories that

treat parts of speech as atomic primitive units. In that respect, decomposition of

categories into features is the opposite of what Radical Construction Grammar

proposes.

On the other hand, features for defining syntactic categories are—or should

be—invoked in order to account for distributional facts about the words or

phrases to which the features are attached. Even a feature like [+N], which appears

to be a further reduction of syntactic structure, is—or should be—invoked when

there is some construction which defines a distribution pattern that permits the

parts of speech that are [+N] but disallows the parts of speech that are [-N]. To

the extent that features specify occurrence/nonoccurrence of the element with the

feature in a construction, the employment of features is in the spirit of Radical

Construction Grammar.

I contend that if one takes all distributional patterns seriously, then one would


end up with a set of features each of which in essence indexed whether or not the

word (or phrase) in question occurs in each role in each construction. For each

mismatch of distributions across constructions, the mismatching words would

somehow have to be indexed with a feature that in effect blocks its occurrence in

one construction or permits its occurrence in the other construction. For instance,

one might attach a feature to the representation of the Object of the English Verb

weigh that permits it to occur without a Preposition (on the assumption that it is

an anomalous Oblique), or prevents it from being passivized (on the assumption

that it is an anomalous Direct Object).

The end result of this process would be a notational variant of Radical Con-

struction Grammar, in which the mapping between words and constructions is

represented by the values of the features indicating the distribution across con-

structions of each word or other grammatical unit. For example, the Object of

weigh would, among other things, have the features which essentially reduce to

<weigh <obj = +ActiveDirectObject, -PassiveSubject>>. Although this feature

representation is one way of modeling Radical Construction Grammar, it has the

drawback that the feature does not separately represent the construction and its

relevant role; the feature label simultaneously encodes both.

A similar argument can be applied to a categorial grammar representation of

the facts. In categorial grammar, syntactic categories are defined in terms of their

combinability with other units to produce a larger unit. For instance, a transitive

verb is defined as something that can combine with a noun phrase to produce a

verb phrase. The category Transitive Verb is notated VP/NP. The categorial

grammar notation also captures distributional relations between words and con-

structions. The categorial grammar representation X/Y means that a word of

category X/Y can combine with element Y in construction X.

Again, I contend that if one takes all distributional patterns seriously, the end

result would be a set of categories that indicates the occurrence of each word in

each construction. To take the weigh example again, one would end up specifying

weigh as a word which can combine with a prepositionless Noun Phrase to its

right to yield an Active Verb Phrase (i.e. weigh is of type ActiveVP/NP), but cannot

combine with the Passive suffix (or with the Passive Auxiliary) to produce a Passive

Verb Phrase (i.e. it is not of type PassiveVP/PassivePrt). Although categorial

grammar is another way of modeling Radical Construction Grammar, it has the

drawback that all combinations must be represented as binary, whereas many

constructions have more than two elements. (Categorial grammar can represent

a construction with more than two elements, but only as a series of binary

combinations.)

1.6.3. Doesn’t RCG create a hopeless proliferation of categories?

How do you label them all?

Radical Construction Grammar does appear to create a proliferation of categories,

but since categories are not atomic primitive elements, this is not a theoretical


problem. The linguist is, in fact, largely free to label the categories as s/he wishes.

Naming the categories is, however, a practical problem for a linguist describing a

language.

The same question arises with respect to Dryer’s argument against universal

categories and relations. In Dryer’s view, all categories are language-specific. Dryer

argues that what to label a category in a language is purely terminological; I will

use the word to describe a name for a category (or construction) that has

no theoretical significance in itself.

Dryer does suggest that using labels such as Noun or Verb instead of X and Y

for, say, Kutenai word classes is convenient: ‘Such a choice of label makes it easier

to remember the labels and to follow discussions of the language, and it does draw

attention to the similarities between these word classes and word classes in other

languages’ (Dryer b: ).

One can suggest a general mnemonic for labeling language-particular cat-

egories: Language + Category, such as Kutenai Verb and English Verb. These labels

can be treated on analogy to proper names for people, as in Surname + Given

Name. The category labels are like given names such as Bill. The fact that the name

Verb occurs in Kutenai Verb and English Verb does not mean that they are

instances of a universal grammatical category Verb with theoretical content, just

as the fact that the name Bill occurs in Bill Croft and Bill Clinton does not mean

that these individuals belong to a unified category with conceptual content.

The labeling problem arises even more acutely in Radical Construction

Grammar. But the naming of elements in particular constructions is just a ter-

minological problem, and a similar mnemonic can be applied here as well. The

simplest mnemonic is for categories defined by the distribution in a role in a single

construction. In this case, the category can be named Language + Construction +Category, such as English Passive Subject. Again, English Passive Subject should

be treated as a proper name for a category; it should not be assumed that the cat-

egory labeled by English Passive Subject is identical to the category labeled English

Active Object, or for that matter to the category labeled English Active Subject—

let alone the category labeled Japanese Passive Subject. Labeling of categories

defined cross-constructionally is not so easy, and I have no simple solution to offer

here. However, many typological universals only require construction-specific cat-

egory labels.

Constructions are also language-specific. In Part III, I will argue that there are

no universal constructions. The labeling problem thus arises here as well, but the

mnemonic solution is straightforward: Language + Construction, as in English

Passive. Again, English Passive is a proper name, and should not be assumed to

be the name of a universal construction identical to, for example, the construc-

tion labeled Japanese Passive.

One advantage of the purely mnemonic labeling of categories and construc-

tions in languages is that a Radical Construction Grammar description or analy-

sis of a particular language can use the terminology already conventionally


established for that language without any theoretical consequences. For example,

a linguist using Radical Construction Grammar for describing or analyzing

Philippine languages can use the traditional descriptions of the verb forms (Actor

Focus, Goal Focus, etc.), even though the terms Goal, Focus, etc. are used to

describe quite different categories in other languages. All that must be remem-

bered is that the labels do not carry any theoretical significance.

One terminological convention that must be adhered to is the orthographic dis-

tinction between the (capitalized) names for grammatical constructions and cat-

egories in a particular language on the one hand and the (lower case) names for

cross-linguistically valid semantic, pragmatic and discourse-functional categories

on the other. For example, it is essential to differentiate the Philippine language

construction name Focus (capitalized) from the pragmatic category of focus

(lower case), since not only are they different types of categories, the grammat-

ical category may not be an expression of the pragmatic category of the same

name.

Although categories (and constructions) are language-specific as morphosyn-

tactic structures, categories and constructions may be compared across languages

according to their function (meaning, in the broad sense defined in §..). The

formulation of cross-linguistic universals is in fact dependent on identifying

categories and constructions across languages in terms of shared function (see

Greenberg b: ; Keenan and Comrie /: ; Croft b: –). For

this reason, when comparing categories and constructions across languages, I will

use lower-case names such as ‘relative clause’. This name should be read as ‘the

category/construction in a language that encodes the function named by this

term’. These lower-case names do have theoretical significance (hence they are

names, not labels).

In discussing language-specific instantiations of these categories and construc-

tions, I will label the construction with the capitalized form of the name, for

example ‘English Relative Clause’. The assertions that are made about the

language-particular categories and constructions labeled in this way should be

interpreted as applying only to the use of the category/construction encoding the

function with which the name is associated. For example, in a discussion of rela-

tive clauses across languages and the English Relative Clause, the lower-case name

denotes ‘the construction encoding the function of a proposition modifying a ref-

erent’ (Keenan and Comrie : ), and any generalization made in the discus-

sion about the grammatical properties of the English Relative Clause construction

apply only to the English Relative Clause construction used for the relative clause

function.

1.6.4. If categories are defined relative to constructions,

how do you identify constructions?

This is the most frequently asked question about Radical Construction Grammar.

Constructions can be identified even if they are primitive units, that is, not defined


in terms of some other theoretical construct. Even though constructions are

primitive units, they are complex (§..) and they have many properties that iden-

tify them. The complex wholes that linguists study (and children learn; see §..)

are instances of constructions, that is, utterances. The parts are the words and

morphemes making up each utterance.

The identification of constructions is essentially a problem,

that is, categorizing the utterances one hears or records into discrete types. In fact,

this is the same problem faced by a field linguist with a corpus of the language

being described, or for that matter the language user and language learner (see

§..). What occurs in natural discourse are constructions, that is, complex syn-

tactic units; we do not hear individual words with category labels attached to

them. Utterances are instances of constructions. A corpus is a set of utterances

which can be sorted into different types according to their grammatical proper-

ties. In other words, from the points of view of the language analyst, language

user, and language learner, the larger units come first.

Categorization of constructions is as easy—or as difficult—as categorization of

any other entities. Constructions often do fall into discrete types. There are dis-

continuities in the corpus of sentences: constructions have distinctive properties

and their elements define distinctive distribution classes. For example, there are

significant discontinuities between the structure of an English Transitive Active

Clause and a Passive Clause, so that the two can be reliably separated. One such

discontinuity is that the Passive Clause lacks a Direct Object which the counter-

part Transitive Active Clause has. There are also other important cues for the

categorization of constructions (see §..). Many constructions involve some

unique combination of substantive morphemes, such as the English Passive con-

struction’s combination of be, Past Participle Verb form (usually an affix), and by

(when the agent phrase occurs).

Finally, and perhaps most important of all, constructions are symbolic units.

The semantics of a construction plays a significant role in differentiating con-

structions for the purpose of categorization and identification. The semantics of

the participant roles encoded by the Active Subject and the Passive Subject are

very different, despite the grammatical similarities of the Subject role in the two

constructions and the identity of the counterpart verb stem. Since constructions

are symbolic units (§..), categorization of utterances into construction types

utilize properties of both form and meaning in constructions.

Of course, categorization is not always as simple as I have made it out to

be here. The English Transitive Active–Passive example represents a level of

schematicity in constructions that has a relatively high - (Rosch

): a relatively high number of properties shared by the members with the cat-

egory plus a relatively low number of properties shared by members across the

categories. Constructions form a taxonomy (§..), and more general and more

specific construction types will have a lower cue-validity that the Transitive

Active/Passive distinction. Also, construction types in other domains of grammar


may not have as high a cue-validity as the example I have used here. Moreover,

research in categorization has demonstrated that different members of a category

have different status in the category: some are more members and

others are more members (Rosch ; see §§..–..; and see

Taylor for a recent survey).

It is true that the variation in the internal structure of categories, and their

discreteness from other categories, is problematic for a linguist analyzing a par-

ticular language, and I have not given any detailed principles for categorization

of constructions in this section. But I must emphasize again that these facts are

problematic to the same extent for the language speaker and language learner. The

best way to understand how to identify constructions is to use the results of

psychological research into categorization and the formation of taxonomies, a

domain that regrettably will not be discussed in detail in this book (see §. and

§..).

1.6.5. How do you capture generalizations for categories across

constructions in RCG?

In order to answer this question, we must first look at how categories are repre-

sented in reductionist theories. To simplify comparison, we will compare reduc-

tionist construction grammar to Radical Construction Grammar.

The syntactic structure of any complex construction is made up of elements

(§..). These elements may be constructions themselves. For example, the

Intransitive Clause construction illustrated in has as a part the Noun Phrase

construction:

() [ [The girls]Noun Phrase [sang]Verb]Intransitive Clause

Ultimately, a complex construction will be broken down into its atomic ele-

ments—words and morphemes—which form categories. It is the representation

of these categories that differentiates a nonreductionist grammatical theory such

as Radical Construction Grammar from reductionist theories.

Both reductionist and nonreductionist syntactic theories assume the existence

of a meronomic (part–whole) structure to complex constructions. The

part–whole relation is indicated implicitly by the nesting of bracketed/boxed ele-

ments in the representation of the morphosyntactic structure of a complex con-

struction. For example, in () the part–whole relation between The girls and The

girls sang is represented implicitly in the nesting of the Noun Phrase brackets

inside the Intransitive Clause brackets. I will refer to this part–whole relation as

an meronomic relation.

Reductionist theories of syntax allow for an element—i.e. a category—to be

part of more than one construction. For example, the part of the Intransitive Con-

struction [S V] labeled ‘Verb’ is also assumed to be a part of the Transitive

Construction as well. If so, the representation of the Verb construction as an inde-

pendent node in the construction network is required. This independent node is


In the representation in Figure ., we will say that there is an mero-

nomic link (labeled ‘m’) between each construction that has a Verb element as a

part on the one hand, and the independent construction representing the Verb

element on the other. Thus, there are at least two different types of links in the

construction network in reductionist construction grammar: taxonomic links

between constructions, discussed in §.., and external meronomic links from

parts of constructions to parts of other constructions, illustrated in Figure ..

In a Radical Construction Grammar representation of a construction, the cat-

egories of the construction are defined by the construction itself. Hence the cat-

egories are unique to each construction. The representations of the Intransitive

construction and the Transitive construction in a reductionist syntactic theory

(such as a reductionist construction grammar model) and in Radical Construc-

tion Grammar are given in Table ..


Figure .. A reductionist construction grammar representation of the category Verb

across constructions

Table .. The representation of constructions in reductionist and nonreductionist

syntactic theories

Reductionist construction Radical Construction

grammar Grammar

Intransitive construction [S V] [IS IV]

Transitive Active construction [S V O] [TS TV TO]

then represented as playing a role in several other constructions. This analysis is

represented in Figure ..

The Radical Construction Grammar representation captures the fact that the

distributional categories defined by the roles in the Transitive construction are not

identical to those defined by the roles in the Intransitive construction. For

example, the class of “Verbs” that can occur in the Transitive construction is not

the same as the class of “Verbs” that can occur in the Intransitive construction.

Hence the two categories are labeled “Intransitive Verb” and “Transitive Verb”. (In

fact, they could be named “Rosencrantz” and “Guildenstern” as far as syntactic

theory is concerned; but I will follow the labeling mnemonic proposed in §...)

However, the Radical Construction Grammar representation does not capture the

fact that Transitive Verbs are inflected with the same inflectional affixes as Intran-

sitive Verbs, or that Transitive Subjects occur in the same position and in the same

pronominal form as Intransitive Subjects.

In a reductionist theory, this fact is captured by describing the categories in the

two constructions as the same category, for example Verb. That is, the Intransi-

tive construction and the Transitive construction share one of their parts, namely,

the Verb, as in Figure .. However, the reductionist representation does not

capture the fact that not all Verbs that occur in the Intransitive construction can

occur in the Transitive construction, or vice versa. Needless to say, an adequate

syntactic theory must be able to capture both the fact that Transitive Verbs and

Intransitive Verbs are distinct classes, and the fact that they share the same inflec-

tions.

The solution to this problem in Radical Construction Grammar can be mo-

tivated by considering again the process of identifying constructions described in

§... Categorizing utterances as instances of constructions is one way of

abstracting away from the input. The result of this process is a network of taxo-

nomic relations among constructions. Another process that occurs in categoriz-

ing utterances is the analysis of utterances breaking them into their constituent

parts. But the analysis of utterances into their component parts is another way of

abstracting from the input. The learner (or analyst) abstracts from two or more

constructions which have parts that share something in common, say, a set of

inflections.

Radical Construction Grammar represents analysis as abstraction from parts

of constructions directly. That is, Radical Construction Grammar allows for taxo-

nomic relations among parts of different constructions. Thus, the Transitive Verb

category in the Transitive construction and the Intransitive Verb category in the

Intransitive construction can be subsumed under a more general category, which

I will call Morphological Verb.

It is absolutely essential to recognize that the commonalities across the subcat-

egories found in various constructions must themselves be justified linguistically.

For example, the justification for a category subsuming Intransitive Verb and

Transitive Verb comes in from the occurrence of the Morphological Verb category

in another construction, namely the morphological construction of Tense-

Agreement (TA) inflection. This is why I have labeled the category Morphologic-

al Verb—in order to make clear that I am not positing a global category Verb

for English (or any other language). There is no global category Verb; just over-

lapping XVerb categories defined by various constructions. The representation

of this subnetwork of verbal categories in Radical Construction Grammar is

given in Figure . (p. ; t is used to indicate that the relations between elements

is taxonomic).

In other words, no schematic syntactic category is ever an independent unit of

grammatical representation in Radical Construction Grammar. Every schematic


category is a part of a construction, which defines that category. Only words, being

completely substantive, may be independent units of grammatical representation

in Radical Construction Grammar. Yet even words are often not analyzable as

wholly independent grammatical units. Words often exist only embedded in con-

structions, such as morphological inflections (see Bybee , chapter ) or syn-

tactic constructions. Words also vary in meaning depending on the constructions

in which they appear (see Chapter and Croft a).

The Radical Construction Grammar representation of categories allows us to

eliminate external meronomic links. In Radical Construction Grammar, the only

type of links between constructions and elements of constructions are taxonomic

links. In this respect, Radical Construction Grammar is a simpler model of gram-

matical knowledge than reductionist construction grammar. In other words, all

types of grammatical generalizations are represented as taxonomic generalization,

that is, categorization (see §..).

Figure . gives a Radical Construction Grammar representation of part of the

taxonomic hierarchy in Figure ..


Figure .. Radical Construction Grammar representation of verbal categories

Figure .. Radical Construction Grammar representation of construction taxonomy

The taxonomic hierarchy in Figure . describes only the taxonomy of argu-

ment structure constructions, that is, constructions representing the encoding of

arguments of predicates (see Fillmore and Kay ; Goldberg ; Croft in

preparation). In Figure ., each construction part is given its own category

label. The construction parts are linked in taxonomies by broken lines, and the

constructions as a whole are linked in taxonomies by heavy black lines.

Each construction and category in the taxonomic hierarchy is motivated by a

grammatical fact of English. Different transitive phrases with kick share grammat-

ical properties, specifically, the encoding of the two arguments. The kick construc-

tion thus permits us to establish the grammatical categories Kick and Kick,

standing for the two arguments of kick in its transitive uses. Likewise, the transi-

tive verb constructions with kick and kiss share grammatical properties, namely the

encoding of the two arguments, and allow us to establish the superordinate cate-

gories TrSbj and TrObj. Finally, the Transitive and Intransitive constructions of

English share grammatical properties, namely the encoding of the one argument

of Intransitive Verbs and one of the arguments of Transitive Verbs (see Chapter ).

The existence of the Clause construction allows us to establish the superordinate

categories SbjArg (“subject as an argument”) and Pred. Note that the Clause

construction does not specify anything about the TrObj element in the Transitive

construction. This is because the Clause construction represents only the general-

ization about the Subject arguments of Transitive and Intransitive clauses.

The actual existence of any of these construction schemas is an empirical ques-

tion (see §..). Although the idiosyncrasy of idioms such as kick the bucket and

kick the habit, and the argument structures of verbs such as kick and kiss, are a

sufficient condition for positing the existence of these specific constructions, the

positing of more general constructions is not a necessary consequence. Likewise,

the elimination of more specific constructions whose properties can be predicted

by more general constructions (such as a semantically regular kick sentence like

She kicked me) is not a necessary consequence. Individual speakers may vary as to

what grammatical generalizations they do or do not form. The positing of more

schematic constructions, and the elimination of more specific constructions with

predictable grammatical properties, is an empirical question, which can only be

answered by the principles discovered by the usage-based model (see §..).

1.6.6. How can a child acquire a grammar without atomic primitive categories?

The essential structure of a radical construction grammar is the many-to-many

mapping between words and constructions (more precisely, between words and

the roles in constructions). This structure is not impossible to store or to learn

(pace Radford : ). Of course, the more constraints we can impose on the

many-to-many mapping, the easier the grammar is to store or learn. But con-

straining the many-to-many mapping such that words in a single class have iden-

tical distributions, as implied by the standard syntactic theories, is empirically

untenable.


Radical Construction Grammar argues that categories and relations are

construction-specific. In addition, Radical Construction Grammar argues that

constructions are language-specific (see Part III). If so, then virtually all of the

formal syntactic properties of grammar are language-specific and therefore must

be learned . The inductive learning process is essentially the same

process described in the preceding section. A child is exposed to utterances in

context—thus, a significant part of the meaning of the utterance is available to

the child from the context of use. The child acquires constructions by categoriz-

ing utterances into their types, in terms of the grammatical properties of the utter-

ances that the child is able to perceive. (It is possible that the child does not

perceive all relevant properties of the input: s/he may not be able to keep track of

inflections, for instance, and so the inflections in the input would not be part of

the construction type s/he induces.)

In the process of categorizing utterances into constructions, the child has effec-

tively induced the categories of the elements of those utterances, since those cat-

egories are defined by the constructions (compare Figure . above). It makes no

sense to ask if children have the same categories as adults do in Radical Con-

struction Grammar, because such a question assumes the existence of global syn-

tactic categories. Categories are defined by constructions, and children do not

have the same constructions as adults do. Eventually, as the child becomes able to

process all aspects of the input, and is exposed to more and more varied instances

of constructions, s/he gradually builds up a taxonomic network of constructions

and their categories that comes to equal that possessed by adult speakers of the

language.

In fact, this model of syntactic acquisition has much empirical evidence in

support of it. A number of recent studies that carefully document in detail chil-

dren’s earliest multiword utterances demonstrate that children start with very nar-

rowly defined constructions and only very gradually expand their production with

different combinations of words and inflectional forms (Braine ; Tomasello

, ; Lieven et al. ; Tomasello et al. ; Pine and Lieven ; Pine et

al. ; Rubino and Pine ; Gathercole et al. ; see Croft and Cruse in prepa-

ration, chapter for a brief summary of their results). This evidence implies a

very gradual inductive process, beginning with very specific constructions and

very gradually inducing more schematic constructions and producing novel utter-

ances with those schemas.

1.6.7. How does RCG relate to other versions of construction grammar?

Radical Construction Grammar, like other versions of construction grammar, uses

a uniform generalized concept of a grammatical construction as a symbolic unit,

atomic or complex, from completely substantive to completely schematic (§..).

Thus, Radical Construction Grammar embraces the syntax–lexicon continuum.

Radical Construction Grammar also uses the same principles of the taxonomic

organization of constructions as most other construction grammar theories


(§..). In particular, Radical Construction Grammar allows for the polysemy

(radial category structure) of some construction types, metaphorical extensions

of constructions, and accepts the usage-based model of the representation of

grammatical information in the taxonomic network of constructions.

Radical Construction Grammar differs from other versions of construction

grammar in the meronomic structure of constructions, that is, the internal

structure of constructions (§..). Radical Construction Grammar retains the

symbolic relations of the parts of constructions as in other construction gram-

mar models, in particular Cognitive Grammar (Langacker : –). In

Radical Construction Grammar, categories are defined by the roles they play

in constructions. Hence the categories are unique to each construction. Radical

Construction Grammar rejects meronomic links between parts of different con-

structions and instead uses taxonomic links between elements of different con-

structions to capture generalizations across categories in different constructions

(§..).

Also, Radical Construction Grammar does not posit syntactic relations between

elements in constructions (see Part II); it posits only semantic relations and sym-

bolic relations (between elements of constructions and the corresponding com-

ponents of the semantic structure of the construction). Finally, as will be argued

in Part III, Radical Construction Grammar argues that constructions are language

specific (Part III), a position not held by all construction grammarians (cf. Kay

and Fillmore : ).

1.6.8. How can I use RCG for grammatical description if all categories are

construction-specific and constructions are language-specific?

Typical reference grammars have chapters describing lexical categories of the lan-

guage, such as the parts of speech, and also chapters describing syntactic roles in

the clause such as subject and object (or ergative and accusative, or actor and

undergoer), and syntactic roles in the phrase such as different types of modifiers

and complements of the head noun.

From a Radical Construction Grammar perspective this is a misleading way to

describe a human language. In a Radical Construction Grammar description of a

language, every chapter would describe a class of constructions of the language.

Instead of parts of speech, chapters would describe the propositional act con-

structions of the language (Chapter ; Croft b). Instead of syntactic roles,

chapters would describe the argument linking constructions of the language

(Chapter ; see Levin , Part One; Goldberg ). Since the generalized

concept of construction includes morphological structures, the description of a

language’s morphology in Radical Construction Grammar would also be in con-

structional terms.

A Radical Construction Grammar description of a language would also

describe in some detail the possible fillers of each role in each construction, since

category labels in themselves do not provide any significant information about


possible fillers. Many descriptions of constructions in traditional grammars do

not give a sufficiently detailed description of the behavior of different lexical

semantic classes across the constructions of the language. Yet substantial differ-

ences in grammatical behavior are found within lexical subclasses. More fine-

grained distinctions than simply ‘verbs’ or ‘count nouns’ are necessary. Previous

studies of lexical classes relevant to various types of constructions include Croft

(: –) for propositional act constructions; Wierzbicka (, Chapter ) for

countability constructions; Craig () and Croft (b) for gender/animacy/

classifying constructions; Herskovits () for spatial relation constructions;

Levin (, Part Two) for argument linking constructions; Croft (c) for voice

constructions; Croft (in preparation) for aspectual semantic classes; and Cristo-

faro (, ch. ) for clausal complement constructions.

Finally, a Radical Construction Grammar description of a language will also

describe in some detail the semantics, pragmatics, and discourse function of each

construction, since constructions are symbolic units and the function of a con-

struction is as much a part of linguistic convention as the construction’s form.

This aspect of grammar is as important as the proper description of mor-

phosyntactic constructions and distributions. In particular, the semantic inter-

pretation of the words or phrases filling roles in constructions often varies

depending on the construction (see §.., §.), and conversely the semantics of

the construction varies depending on the class of words or phrases filling the con-

structional roles (see §.., §..). These differences in semantic interpretation

are a central part of grammatical description.

Constructions are language-specific in their morphosyntactic properties, but

their function in structuring and communicating information is not. A Radical

Construction Grammar description of a grammar would organize constructions

in terms of the functions that they perform. The organization of constructions by

functions will in fact not be that different from common practice in reference

grammars, because the traditional organization of grammatical descriptions is

actually based on function as much as on form.

Radical Construction Grammar essentially allows the descriptive linguist to

describe the structure of a language in the language’s own terms, since construc-

tions are language-specific. The functional organization of a grammatical descrip-

tion will allow the description to be easily followed by readers unfamiliar with the

language. Finally, the typological generalizations presented in this book and in the

monographs and reference works cited herein will allow the field linguist to place

the constructions of her or his language in a wider grammatical context.

1.6.9. If categories are construction-specific and constructions are

language-specific, does this mean that there is no Universal Grammar,

and no language universals?

If Radical Construction Grammar argues that virtually all of syntactic structure

is language-particular, and reduces that structure to the absolute minimum on


top of that, then are there any universals of language left? Is there no Universal

Grammar?

In one sense, there is no Universal Grammar. That is, there is no universal syn-

tactic template to which the grammars of all particular languages conform. Nor

is there an inventory of universal syntactic categories, relations or even construc-

tions which the grammars of all particular languages draw from. This is in fact a

good thing, because the endless cycle of syntactic theories witnessed in the last

century is a consequence of the futile search for the mirage of the “right” univer-

sal template or set of universal categories, relations, and syntactic structures.

There are universals of language, but not in syntactic structure taken by itself.

The universals of language are found in semantic structure and in symbolic struc-

ture, that is, the mapping between linguistic function and linguistic form. In

Chapter , I will introduce the notion of a conceptual space, now gaining wider

use among typologists, and the principles of typological markedness that govern

the form–function mapping. In the following chapters, these theoretical con-

structs will be used to form positive hypotheses of language universals. Many of

these hypotheses are not new, having been proposed in typological research over

the past four decades. However, Radical Construction Grammar allows these

results to take their rightful place in the context of a cross-linguistically valid

theory of syntax.

1.7. Conclusion and Prospects

Radical Construction Grammar has a number of advantages over reductionist

syntactic theories. It avoids the inconsistencies and self-fulfilling analyses of

methodological opportunism in using the distributional method. There is no need

to ignore, dismiss, or treat as exceptional distributional facts that do not fit the

global categories posited by reductionist theories. It allows for the proper focus

of attention on the richness and diversity of a speaker’s knowledge of the con-

structions of his/her language. It also allows one to capture both the differences

and similarities among categories defined by different constructions in the same

language, as well as across languages.

The mismatches in distribution within and across languages do fall into sys-

tematic patterns. These patterns have been analyzed by typologists and include

such phenomena as implicational hierarchies, typological prototypes, grammati-

calization paths, etc. Not only do reductionist syntactic theories founder on the

mismatches; their models of syntactic representation often do not capture the pat-

terns that the mismatches conform to.

It should be noted that the same arguments against reductionist theories of

syntactic representation also apply to reductionist theories of phonological

and semantic representation. In phonology, there are problems in defining ‘vowel’

vs. ‘consonant’, deciding whether certain syllables exist or not as such for certain

prosodic processes (‘extrametricality’), and even in defining ‘segment’ and


‘syllable’ from a phonetic point of view. In a nonreductionist phonological theory,

phonetically specified word forms and schematic phonotactic (and prosodic)

templates generalized from them would be the representational primitives, and

syllable and segment categories would be derivative.

In semantics, distributional analysis is used to identify semantic categories (see

e.g. Cruse ). Not surprisingly, problems arise in defining various sorts of

semantic categories, and even such basic concepts as identity and distinctness of

word senses. In a nonreductionist semantic theory, complex semantic structures

such as frames and the complex semantic structures found in constructions would

be the representational primitives, and the semantic categories of components of

semantic frames and other complex semantic structures would be derivative.

The development of a Radical Templatic Phonology and a Radical Frame

Semantics should be a priority for nonreductionist linguistic theory. This book

will not present either theory. However, an important aspect of syntactic repre-

sentation is the counterpart semantic interpretation. In construction grammar, a

construction is symbolic: it is a pairing of a morphosyntactic structure with a

semantic structure. Hence we cannot entirely avoid the question of semantic rep-

resentation, although this book focuses on syntactic representation. In particular,

one cannot discuss cross-linguistically valid criteria for syntactic representation

without addressing the cross-linguistic validity of semantic representation; this is

done in Chapter .

I assume that the semantic structures in particular constructions function as

units; these units would be the primitives of Radical Frame Semantics. These units

can be related to one another in a conceptual space, as noted in §..; conceptual

spaces are described in §. and will be used throughout Parts I and III of this

book. In Part II, on the internal structure of constructions and their meanings, I

make minimal assumptions about the semantic structure of the conventional

meaning of a construction. I assume only that a semantic structure is made up of

components and relations among those components (see also §..), and that

morphosyntactic elements can denote (symbolize) either semantic components

or semantic relations. Hence, the semantic representations in Part II merely indi-

cate semantic components and their relations. Proposing and defending more

precise semantic representations would require another volume.

In fact, one of the reasons why this book is half the length of its nearest coun-

terparts—Givón , ; Langacker , a; Van Valin and LaPolla —is

that the latter volumes (especially Langacker’s) discuss semantic issues at greater

length than I do here. Nevertheless, it is possible to discuss—and resolve—fun-

damental issues in syntactic theory without presenting a full-fledged semantic

theory, in fact by making only the minimal assumptions presented in the two pre-

ceding paragraphs. That is what this book seeks to achieve.


p i from syntactic categories to semantic...

Documents